Gótico americano na paleta da Mona Lisa: reorganize os pixels


377

Você recebe duas imagens em cores verdadeiras, a Fonte e a Paleta. Eles não têm necessariamente as mesmas dimensões, mas é garantido que suas áreas são iguais, ou seja, têm o mesmo número de pixels.

Sua tarefa é criar um algoritmo que faça a cópia de aparência mais precisa da Origem usando apenas os pixels na Paleta. Cada pixel na Paleta deve ser usado exatamente uma vez em uma posição exclusiva nesta cópia. A cópia deve ter as mesmas dimensões que a Origem.

Este script Python pode ser usado para garantir que essas restrições sejam atendidas:

from PIL import Image
def check(palette, copy):
    palette = sorted(Image.open(palette).convert('RGB').getdata())
    copy = sorted(Image.open(copy).convert('RGB').getdata())
    print 'Success' if copy == palette else 'Failed'

check('palette.png', 'copy.png')

Aqui estão várias fotos para teste. Todos eles têm a mesma área. Seu algoritmo deve funcionar para quaisquer duas imagens com áreas iguais, não apenas o American Gothic e a Mona Lisa. Obviamente, você deve mostrar sua saída.

gótico americano Monalisa Noite estrelada O grito Rio arco Iris

Obrigado à Wikipedia pelas imagens de pinturas famosas.

Pontuação

Este é um concurso de popularidade, para que a resposta mais votada ganhe. Mas tenho certeza de que há muitas maneiras de ser criativo com isso!

Animação

Millinon teve a ideia de que seria legal ver os pixels se reorganizarem. Eu também achava que escrevi esse script Python que pega duas imagens feitas da mesma cor e desenha as imagens intermediárias entre elas. Atualização: Acabei de revisá-lo para que cada pixel mova a quantidade mínima necessária. Não é mais aleatório.

Primeiro, a Mona Lisa se transforma no gótico americano do aditsu. Em seguida, o gótico americano de bitpwner (da Mona Lisa) se transforma em aditsu. É incrível que as duas versões compartilhem exatamente a mesma paleta de cores.

Mona Lisa para animação gótica americana animação entre duas versões do gótico americano feito de Mona Lisa

Os resultados são realmente surpreendentes. Aqui está o arco-íris de aditsu, Mona Lisa (diminuiu a velocidade para mostrar detalhes).

esferas do arco-íris para animação Mona Lisa

Esta última animação não está necessariamente relacionada ao concurso. Ele mostra o que acontece quando meu script é usado para girar uma imagem 90 graus.

animação de rotação de árvore


22
Para aumentar os resultados da sua pergunta, considere habilitá-la: "Gótico americano na paleta da Mona Lisa: reorganize os pixels"
9609

14
Olá, só quero parabenizá-lo por esse desafio original! Muito refrescante e interessante.
bolov 9/07/2014

6
Fico feliz que isso não seja um [código-golfe].
Ming-Tang

13
Meu limite de dados móveis sofre uma terrível queimadura toda vez que visito esta página.
Vetorizado

Respostas:


159

Java - GUI com transformação aleatória progressiva

Eu tentei muitas coisas, algumas delas muito complicadas, e finalmente voltei a esse código relativamente simples:

import java.awt.BorderLayout;
import java.awt.event.ActionEvent;
import java.awt.event.ActionListener;
import java.awt.image.BufferedImage;
import java.io.File;
import java.io.IOException;
import java.util.Random;

import javax.imageio.ImageIO;
import javax.swing.ImageIcon;
import javax.swing.JButton;
import javax.swing.JFrame;
import javax.swing.JLabel;
import javax.swing.Timer;

@SuppressWarnings("serial")
public class CopyColors extends JFrame {
    private static final String SOURCE = "spheres";
    private static final String PALETTE = "mona";
    private static final int COUNT = 10000;
    private static final int DELAY = 20;
    private static final int LUM_WEIGHT = 10;

    private static final double[] F = {0.114, 0.587, 0.299};
    private final BufferedImage source;
    protected final BufferedImage dest;
    private final int sw;
    private final int sh;
    private final int n;
    private final Random r = new Random();
    private final JLabel l;

    public CopyColors(final String sourceName, final String paletteName) throws IOException {
        super("CopyColors by aditsu");
        source = ImageIO.read(new File(sourceName + ".png"));
        final BufferedImage palette = ImageIO.read(new File(paletteName + ".png"));
        sw = source.getWidth();
        sh = source.getHeight();
        final int pw = palette.getWidth();
        final int ph = palette.getHeight();
        n = sw * sh;
        if (n != pw * ph) {
            throw new RuntimeException();
        }
        dest = new BufferedImage(sw, sh, BufferedImage.TYPE_INT_RGB);
        for (int i = 0; i < sh; ++i) {
            for (int j = 0; j < sw; ++j) {
                final int x = i * sw + j;
                dest.setRGB(j, i, palette.getRGB(x % pw, x / pw));
            }
        }
        l = new JLabel(new ImageIcon(dest));
        add(l);
        final JButton b = new JButton("Save");
        add(b, BorderLayout.SOUTH);
        b.addActionListener(new ActionListener() {
            @Override
            public void actionPerformed(final ActionEvent e) {
                try {
                    ImageIO.write(dest, "png", new File(sourceName + "-" + paletteName + ".png"));
                } catch (IOException ex) {
                    ex.printStackTrace();
                }
            }
        });
    }

    protected double dist(final int x, final int y) {
        double t = 0;
        double lx = 0;
        double ly = 0;
        for (int i = 0; i < 3; ++i) {
            final double xi = ((x >> (i * 8)) & 255) * F[i];
            final double yi = ((y >> (i * 8)) & 255) * F[i];
            final double d = xi - yi;
            t += d * d;
            lx += xi;
            ly += yi;
        }
        double l = lx - ly;
        return t + l * l * LUM_WEIGHT;
    }

    public void improve() {
        final int x = r.nextInt(n);
        final int y = r.nextInt(n);
        final int sx = source.getRGB(x % sw, x / sw);
        final int sy = source.getRGB(y % sw, y / sw);
        final int dx = dest.getRGB(x % sw, x / sw);
        final int dy = dest.getRGB(y % sw, y / sw);
        if (dist(sx, dx) + dist(sy, dy) > dist(sx, dy) + dist(sy, dx)) {
            dest.setRGB(x % sw, x / sw, dy);
            dest.setRGB(y % sw, y / sw, dx);
        }
    }

    public void update() {
        l.repaint();
    }

    public static void main(final String... args) throws IOException {
        final CopyColors x = new CopyColors(SOURCE, PALETTE);
        x.setSize(800, 600);
        x.setLocationRelativeTo(null);
        x.setDefaultCloseOperation(JFrame.EXIT_ON_CLOSE);
        x.setVisible(true);
        new Timer(DELAY, new ActionListener() {
            @Override
            public void actionPerformed(final ActionEvent e) {
                for (int i = 0; i < COUNT; ++i) {
                    x.improve();
                }
                x.update();
            }
        }).start();
    }
}

Todos os parâmetros relevantes são definidos como constantes no início da classe.

O programa primeiro copia a imagem da paleta para as dimensões de origem, depois escolhe repetidamente 2 pixels aleatórios e os troca se isso os aproximasse da imagem de origem. "Mais próximo" é definido usando uma função de distância da cor que calcula a diferença entre os componentes r, g, b (ponderado por luma) juntamente com a diferença total de luma, com um peso maior para luma.

Leva apenas alguns segundos para que as formas se formem, mas um pouco mais para as cores se unirem. Você pode salvar a imagem atual a qualquer momento. Normalmente, esperava cerca de 1 a 3 minutos antes de salvar.

Resultados:

Diferentemente de outras respostas, todas essas imagens foram geradas usando exatamente os mesmos parâmetros (exceto os nomes dos arquivos).

Paleta gótica americana

mona-gótico grito-gótico

Paleta Mona Lisa

gothic-mona grito-mona esferas-mona

Paleta Noite Estrelada

mona-noite noite de grito esferas-noite

A paleta Scream

grito gótico grito mona grito noturno esferas-gritar

Paleta de esferas

Acho que este é o teste mais difícil e todos devem postar seus resultados com esta paleta:

esferas góticas mona-esferas esferas de grito

Desculpe, não achei a imagem do rio muito interessante, por isso não a incluí.

Também adicionei um vídeo em https://www.youtube.com/watch?v=_-w3cKL5teM , ele mostra o que o programa faz (não exatamente em tempo real, mas semelhante) e mostra o movimento gradual de pixels usando o python de Calvin roteiro. Infelizmente, a qualidade do vídeo é significativamente prejudicada pela codificação / compactação do youtube.


2
@Quincunx E eu não estou chamando invokeLater seja, atirar-me: p Além disso, graças :)
aditsu

16
Melhor resposta até agora ...
Yuval Filmus

8
Em caso de dúvida, força bruta? Parece uma excelente solução, eu adoraria ver uma animação para isso, talvez até um vídeo em vez de um gif.
Lilienthal

3
Você pode estender o algoritmo um pouco para um recozimento completo simulado para uma pequena melhoria. O que você está fazendo já está muito próximo (mas é ganancioso). Encontrar a permutação que minimiza a distância parece um grande problema de otimização, portanto esse tipo de heurística é adequado. @ Lilienthal, isso não é uma força bruta, na verdade é próximo das técnicas de otimização mais usadas.
Szabolcs

3
Esse algoritmo tem os melhores resultados de longe. E é tão simples. Isso torna um vencedor claro para mim.
Leif

118

Java

import java.awt.Point;
import java.awt.image.BufferedImage;
import java.io.File;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Collections;
import java.util.Comparator;
import java.util.HashSet;
import java.util.Iterator;
import java.util.LinkedList;
import java.util.List;
import java.util.Random;
import javax.imageio.ImageIO;

/**
 *
 * @author Quincunx
 */
public class PixelRearranger {

    public static void main(String[] args) throws IOException {
        BufferedImage source = ImageIO.read(resource("American Gothic.png"));
        BufferedImage palette = ImageIO.read(resource("Mona Lisa.png"));
        BufferedImage result = rearrange(source, palette);
        ImageIO.write(result, "png", resource("result.png"));
        validate(palette, result);
    }

    public static class MInteger {
        int val;

        public MInteger(int i) {
            val = i;
        }
    }

    public static BufferedImage rearrange(BufferedImage source, BufferedImage palette) {
        BufferedImage result = new BufferedImage(source.getWidth(),
                source.getHeight(), BufferedImage.TYPE_INT_RGB);

        //This creates a list of points in the Source image.
        //Then, we shuffle it and will draw points in that order.
        List<Point> samples = getPoints(source.getWidth(), source.getHeight());
        System.out.println("gotPoints");

        //Create a list of colors in the palette.
        rgbList = getColors(palette);
        Collections.sort(rgbList, rgb);
        rbgList = new ArrayList<>(rgbList);
        Collections.sort(rbgList, rbg);
        grbList = new ArrayList<>(rgbList);
        Collections.sort(grbList, grb);
        gbrList = new ArrayList<>(rgbList);
        Collections.sort(gbrList, gbr);
        brgList = new ArrayList<>(rgbList);
        Collections.sort(brgList, brg);
        bgrList = new ArrayList<>(rgbList);
        Collections.sort(bgrList, bgr);

        while (!samples.isEmpty()) {
            Point currentPoint = samples.remove(0);
            int sourceAtPoint = source.getRGB(currentPoint.x, currentPoint.y);
            int bestColor = search(new MInteger(sourceAtPoint));
            result.setRGB(currentPoint.x, currentPoint.y, bestColor);
        }
        return result;
    }

    public static List<Point> getPoints(int width, int height) {
        HashSet<Point> points = new HashSet<>(width * height);
        for (int x = 0; x < width; x++) {
            for (int y = 0; y < height; y++) {
                points.add(new Point(x, y));
            }
        }
        List<Point> newList = new ArrayList<>();
        List<Point> corner1 = new LinkedList<>();
        List<Point> corner2 = new LinkedList<>();
        List<Point> corner3 = new LinkedList<>();
        List<Point> corner4 = new LinkedList<>();

        Point p1 = new Point(width / 3, height / 3);
        Point p2 = new Point(width * 2 / 3, height / 3);
        Point p3 = new Point(width / 3, height * 2 / 3);
        Point p4 = new Point(width * 2 / 3, height * 2 / 3);

        newList.add(p1);
        newList.add(p2);
        newList.add(p3);
        newList.add(p4);
        corner1.add(p1);
        corner2.add(p2);
        corner3.add(p3);
        corner4.add(p4);
        points.remove(p1);
        points.remove(p2);
        points.remove(p3);
        points.remove(p4);

        long seed = System.currentTimeMillis();
        Random c1Random = new Random(seed += 179426549); //The prime number pushes the first numbers apart
        Random c2Random = new Random(seed += 179426549); //Or at least I think it does.
        Random c3Random = new Random(seed += 179426549);
        Random c4Random = new Random(seed += 179426549);

        Dir NW = Dir.NW;
        Dir N = Dir.N;
        Dir NE = Dir.NE;
        Dir W = Dir.W;
        Dir E = Dir.E;
        Dir SW = Dir.SW;
        Dir S = Dir.S;
        Dir SE = Dir.SE;
        while (!points.isEmpty()) {
            putPoints(newList, corner1, c1Random, points, NW, N, NE, W, E, SW, S, SE);
            putPoints(newList, corner2, c2Random, points, NE, N, NW, E, W, SE, S, SW);
            putPoints(newList, corner3, c3Random, points, SW, S, SE, W, E, NW, N, NE);
            putPoints(newList, corner4, c4Random, points, SE, S, SW, E, W, NE, N, NW);
        }
        return newList;
    }

    public static enum Dir {
        NW(-1, -1), N(0, -1), NE(1, -1), W(-1, 0), E(1, 0), SW(-1, 1), S(0, 1), SE(1, 1);
        final int dx, dy;

        private Dir(int dx, int dy) {
            this.dx = dx;
            this.dy = dy;
        }

        public Point add(Point p) {
            return new Point(p.x + dx, p.y + dy);
        }
    }

    public static void putPoints(List<Point> newList, List<Point> listToAddTo, Random rand,
                                 HashSet<Point> points, Dir... adj) {
        List<Point> newPoints = new LinkedList<>();
        for (Iterator<Point> iter = listToAddTo.iterator(); iter.hasNext();) {
            Point p = iter.next();
            Point pul = adj[0].add(p);
            Point pu = adj[1].add(p);
            Point pur = adj[2].add(p);
            Point pl = adj[3].add(p);
            Point pr = adj[4].add(p);
            Point pbl = adj[5].add(p);
            Point pb = adj[6].add(p);
            Point pbr = adj[7].add(p);
            int allChosen = 0;
            if (points.contains(pul)) {
                if (rand.nextInt(5) == 0) {
                    allChosen++;
                    newPoints.add(pul);
                    newList.add(pul);
                    points.remove(pul);
                }
            } else {
                allChosen++;
            }
            if (points.contains(pu)) {
                if (rand.nextInt(5) == 0) {
                    allChosen++;
                    newPoints.add(pu);
                    newList.add(pu);
                    points.remove(pu);
                }
            } else {
                allChosen++;
            }
            if (points.contains(pur)) {
                if (rand.nextInt(3) == 0) {
                    allChosen++;
                    newPoints.add(pur);
                    newList.add(pur);
                    points.remove(pur);
                }
            } else {
                allChosen++;
            }
            if (points.contains(pl)) {
                if (rand.nextInt(5) == 0) {
                    allChosen++;
                    newPoints.add(pl);
                    newList.add(pl);
                    points.remove(pl);
                }
            } else {
                allChosen++;
            }
            if (points.contains(pr)) {
                if (rand.nextInt(2) == 0) {
                    allChosen++;
                    newPoints.add(pr);
                    newList.add(pr);
                    points.remove(pr);
                }
            } else {
                allChosen++;
            }
            if (points.contains(pbl)) {
                if (rand.nextInt(5) == 0) {
                    allChosen++;
                    newPoints.add(pbl);
                    newList.add(pbl);
                    points.remove(pbl);
                }
            } else {
                allChosen++;
            }
            if (points.contains(pb)) {
                if (rand.nextInt(3) == 0) {
                    allChosen++;
                    newPoints.add(pb);
                    newList.add(pb);
                    points.remove(pb);
                }
            } else {
                allChosen++;
            }
            if (points.contains(pbr)) {
                newPoints.add(pbr);
                newList.add(pbr);
                points.remove(pbr);
            }
            if (allChosen == 7) {
                iter.remove();
            }
        }
        listToAddTo.addAll(newPoints);
    }

    public static List<MInteger> getColors(BufferedImage img) {
        int width = img.getWidth();
        int height = img.getHeight();
        List<MInteger> colors = new ArrayList<>(width * height);
        for (int x = 0; x < width; x++) {
            for (int y = 0; y < height; y++) {
                colors.add(new MInteger(img.getRGB(x, y)));
            }
        }
        return colors;
    }

    public static int search(MInteger color) {
        int rgbIndex = binarySearch(rgbList, color, rgb);
        int rbgIndex = binarySearch(rbgList, color, rbg);
        int grbIndex = binarySearch(grbList, color, grb);
        int gbrIndex = binarySearch(gbrList, color, gbr);
        int brgIndex = binarySearch(brgList, color, brg);
        int bgrIndex = binarySearch(bgrList, color, bgr);

        double distRgb = dist(rgbList.get(rgbIndex), color);
        double distRbg = dist(rbgList.get(rbgIndex), color);
        double distGrb = dist(grbList.get(grbIndex), color);
        double distGbr = dist(gbrList.get(gbrIndex), color);
        double distBrg = dist(brgList.get(brgIndex), color);
        double distBgr = dist(bgrList.get(bgrIndex), color);

        double minDist = Math.min(Math.min(Math.min(Math.min(Math.min(
                distRgb, distRbg), distGrb), distGbr), distBrg), distBgr);

        MInteger ans;
        if (minDist == distRgb) {
            ans = rgbList.get(rgbIndex);
        } else if (minDist == distRbg) {
            ans = rbgList.get(rbgIndex);
        } else if (minDist == distGrb) {
            ans = grbList.get(grbIndex);
        } else if (minDist == distGbr) {
            ans = grbList.get(grbIndex);
        } else if (minDist == distBrg) {
            ans = grbList.get(rgbIndex);
        } else {
            ans = grbList.get(grbIndex);
        }
        rgbList.remove(ans);
        rbgList.remove(ans);
        grbList.remove(ans);
        gbrList.remove(ans);
        brgList.remove(ans);
        bgrList.remove(ans);
        return ans.val;
    }

    public static int binarySearch(List<MInteger> list, MInteger val, Comparator<MInteger> cmp){
        int index = Collections.binarySearch(list, val, cmp);
        if (index < 0) {
            index = ~index;
            if (index >= list.size()) {
                index = list.size() - 1;
            }
        }
        return index;
    }

    public static double dist(MInteger color1, MInteger color2) {
        int c1 = color1.val;
        int r1 = (c1 & 0xFF0000) >> 16;
        int g1 = (c1 & 0x00FF00) >> 8;
        int b1 = (c1 & 0x0000FF);

        int c2 = color2.val;
        int r2 = (c2 & 0xFF0000) >> 16;
        int g2 = (c2 & 0x00FF00) >> 8;
        int b2 = (c2 & 0x0000FF);

        int dr = r1 - r2;
        int dg = g1 - g2;
        int db = b1 - b2;
        return Math.sqrt(dr * dr + dg * dg + db * db);
    }

    //This method is here solely for my ease of use (I put the files under <Project Name>/Resources/ )
    public static File resource(String fileName) {
        return new File(System.getProperty("user.dir") + "/Resources/" + fileName);
    }

    static List<MInteger> rgbList;
    static List<MInteger> rbgList;
    static List<MInteger> grbList;
    static List<MInteger> gbrList;
    static List<MInteger> brgList;
    static List<MInteger> bgrList;
    static Comparator<MInteger> rgb = (color1, color2) -> color1.val - color2.val;
    static Comparator<MInteger> rbg = (color1, color2) -> {
        int c1 = color1.val;
        int c2 = color2.val;
        c1 = ((c1 & 0xFF0000)) | ((c1 & 0x00FF00) >> 8) | ((c1 & 0x0000FF) << 8);
        c2 = ((c2 & 0xFF0000)) | ((c2 & 0x00FF00) >> 8) | ((c2 & 0x0000FF) << 8);
        return c1 - c2;
    };
    static Comparator<MInteger> grb = (color1, color2) -> {
        int c1 = color1.val;
        int c2 = color2.val;
        c1 = ((c1 & 0xFF0000) >> 8) | ((c1 & 0x00FF00) << 8) | ((c1 & 0x0000FF));
        c2 = ((c2 & 0xFF0000) >> 8) | ((c2 & 0x00FF00) << 8) | ((c2 & 0x0000FF));
        return c1 - c2;
    };

    static Comparator<MInteger> gbr = (color1, color2) -> {
        int c1 = color1.val;
        int c2 = color2.val;
        c1 = ((c1 & 0xFF0000) >> 16) | ((c1 & 0x00FF00) << 8) | ((c1 & 0x0000FF) << 8);
        c2 = ((c2 & 0xFF0000) >> 16) | ((c2 & 0x00FF00) << 8) | ((c2 & 0x0000FF) << 8);
        return c1 - c2;
    };

    static Comparator<MInteger> brg = (color1, color2) -> {
        int c1 = color1.val;
        int c2 = color2.val;
        c1 = ((c1 & 0xFF0000) >> 8) | ((c1 & 0x00FF00) >> 8) | ((c1 & 0x0000FF) << 16);
        c2 = ((c2 & 0xFF0000) >> 8) | ((c2 & 0x00FF00) >> 8) | ((c2 & 0x0000FF) << 16);
        return c1 - c2;
    };

    static Comparator<MInteger> bgr = (color1, color2) -> {
        int c1 = color1.val;
        int c2 = color2.val;
        c1 = ((c1 & 0xFF0000) >> 16) | ((c1 & 0x00FF00)) | ((c1 & 0x0000FF) << 16);
        c2 = ((c2 & 0xFF0000) >> 16) | ((c2 & 0x00FF00)) | ((c2 & 0x0000FF) << 16);
        return c1 - c2;
    };

    public static void validate(BufferedImage palette, BufferedImage result) {
        List<Integer> paletteColors = getTrueColors(palette);
        List<Integer> resultColors = getTrueColors(result);
        Collections.sort(paletteColors);
        Collections.sort(resultColors);
        System.out.println(paletteColors.equals(resultColors));
    }

    public static List<Integer> getTrueColors(BufferedImage img) {
        int width = img.getWidth();
        int height = img.getHeight();
        List<Integer> colors = new ArrayList<>(width * height);
        for (int x = 0; x < width; x++) {
            for (int y = 0; y < height; y++) {
                colors.add(img.getRGB(x, y));
            }
        }
        Collections.sort(colors);
        return colors;
    }
}

Minha abordagem funciona encontrando a cor mais próxima de cada pixel (bem, provavelmente a mais próxima), em três espaços, já que as cores são 3D.

Isso funciona criando uma lista de todos os pontos que precisamos preencher e uma lista de todas as cores possíveis que podemos usar. Nós randomizamos a lista de pontos (para que a imagem fique melhor), depois analisamos cada ponto e obtemos a cor da imagem de origem.

Atualizar: eu costumava simplesmente pesquisar binário, então o vermelho combinava melhor que o verde, que combinava melhor que o azul. Agora mudei para fazer seis pesquisas binárias (todas as permutações possíveis) e, em seguida, escolha a cor mais próxima. Leva apenas 6 vezes mais (1 minuto). Enquanto as fotos ainda estão granuladas, as cores combinam melhor.

Atualização 2: não seleciono mais a lista aleatoriamente. Em vez disso, escolho quatro pontos seguindo a regra dos terços e, em seguida, organizo os pontos aleatoriamente, com preferência para preencher o centro.

Nota: Veja o histórico de revisões para as fotos antigas.

Mona Lisa -> Rio:

insira a descrição da imagem aqui

Mona Lisa -> Gótico Americano:

insira a descrição da imagem aqui

Mona Lisa -> Esferas traçadas por raios:

insira a descrição da imagem aqui

Noite Estrelada -> Mona Lisa:

insira a descrição da imagem aqui


Aqui está um GIF animado mostrando como a imagem foi construída:

insira a descrição da imagem aqui

E mostrando os pixels tirados da Mona Lisa:

insira a descrição da imagem aqui


11
Isso é incrível. Eu não teria pensado isso possível.
AndoDaan

6
Duvido que seria trivial, mas seria incrível poder produzir uma versão animada que mostra os pixels que se deslocam da imagem original para a final.
millinon

2
Eu acho que você não entendeu o problema. Você precisa reorganizar os pixels da paleta para criar a cópia, e não simplesmente usar cores da paleta. Cada cor distinta deve ser usada na cópia exatamente o mesmo número de vezes que apareceu na paleta. Suas imagens não estão passando no meu script.
9788 Calvin's Hobbies

7
@ Quincunx Como se vê, meu script estava correto (embora eu o tenha simplificado para a posteridade) e o mesmo acontece com o seu programa. Por razões que não tenho certeza, a imagem da Mona Lisa mudou ligeiramente quando foi carregada. Notei que o pixel em (177, 377) tinha um rgb de (0, 0, 16) on-line e (0, 0, 14) no meu computador doméstico. Substituí os jpegs por pngs para evitar problemas com um tipo de arquivo com perdas. Os dados de pixel nas imagens não deveriam ter sido alterados, mas pode ser recomendável fazer o download novamente das imagens.
9788 Calvin's Hobbies

8
Esta não deve ser a resposta mais popular. O algoritmo é desnecessariamente complicado e os resultados são ruins, embora pareçam interessantes. Compare a transformação de Mona Lisa em esferas traçadas por raios com o resultado do arditsu
Leif

97

Perl, com espaço de cores no laboratório e pontilhamento

Nota: Agora eu tenho uma solução C também.

Usa uma abordagem semelhante à do aditsu (escolha duas posições aleatórias e troque os pixels nessas posições, se a imagem parecer mais com a imagem de destino), com duas grandes melhorias:

  1. Usa o espaço de cores CIE L a b * de cores para comparar cores - a métrica euclidiana nesse espaço é uma aproximação muito boa à diferença perceptiva entre duas cores, para que os mapeamentos de cores sejam mais precisos que RGB ou HSV / HSL.
  2. Após um passe inicial colocando os pixels na melhor posição única possível, ele faz um passe adicional com um pontilhamento aleatório. Em vez de comparar os valores de pixel nas duas posições de swap, ele calcula o valor médio de pixel de uma vizinhança 3x3 centralizada nas posições de swap. Se uma troca melhorar as cores médias dos bairros, é permitido, mesmo que torne os pixels individuais menos precisos. Para alguns pares de imagens, isso tem um efeito duvidoso na qualidade (e torna o efeito da paleta menos impressionante), mas para alguns (como esferas -> qualquer coisa) isso ajuda bastante. O fator "detalhe" enfatiza o pixel central em um grau variável. Aumentá-lo diminui a quantidade geral de pontilhamento, mas mantém mais detalhes da imagem de destino. A otimização pontilhada é mais lenta,

A média dos valores do laboratório, como o pontilhamento, não é realmente justificada (eles devem ser convertidos para XYZ, com média e convertidos novamente), mas funciona bem para esses propósitos.

Essas imagens têm limites de terminação de 100 e 200 (finalizam a primeira fase quando menos de 1 em 5000 swaps são aceitos e a segunda fase em 1 em 2500) e um fator de detalhe de pontilhamento de 12 (um pouco mais apertado que o conjunto anterior) ) Nessa configuração de super alta qualidade, as imagens levam muito tempo para serem geradas, mas com paralelização todo o trabalho ainda é concluído em uma hora na minha caixa de 6 núcleos. Aumentar os valores até 500 ou mais termina as imagens em alguns minutos, elas parecem um pouco menos polidas. Eu queria mostrar o algoritmo da melhor forma aqui.

Código não é nada bonito:

#!/usr/bin/perl
use strict;
use warnings;
use Image::Magick;
use Graphics::ColorObject 'RGB_to_Lab';
use List::Util qw(sum max);

my $source = Image::Magick->new;
$source->Read($ARGV[0]);
my $target = Image::Magick->new;
$target->Read($ARGV[1]);
my ($limit1, $limit2, $detail) = @ARGV[2,3,4];

my ($width, $height) = ($target->Get('width'), $target->Get('height'));

# Transfer the pixels of the $source onto a new canvas with the diemnsions of $target
$source->Set(magick => 'RGB');
my $img = Image::Magick->new(size => "${width}x${height}", magick => 'RGB', depth => 8);
$img->BlobToImage($source->ImageToBlob);

my ($made, $rejected) = (0,0);

system("rm anim/*.png");

my (@img_lab, @target_lab);
for my $x (0 .. $width) {
  for my $y (0 .. $height) {
    $img_lab[$x][$y] = RGB_to_Lab([$img->getPixel(x => $x, y => $y)], 'sRGB');
    $target_lab[$x][$y] = RGB_to_Lab([$target->getPixel(x => $x, y => $y)], 'sRGB');
  }
}

my $n = 0;
my $frame = 0;
my $mode = 1;

while (1) {
  $n++;

  my $swap = 0;
  my ($x1, $x2, $y1, $y2) = (int rand $width, int rand $width, int rand $height, int rand $height);
  my ($dist, $dist_swapped);

  if ($mode == 1) {
    $dist = (sum map { ($img_lab[$x1][$y1][$_] - $target_lab[$x1][$y1][$_])**2 } 0..2)
          + (sum map { ($img_lab[$x2][$y2][$_] - $target_lab[$x2][$y2][$_])**2 } 0..2);

    $dist_swapped = (sum map { ($img_lab[$x2][$y2][$_] - $target_lab[$x1][$y1][$_])**2 } 0..2)
                  + (sum map { ($img_lab[$x1][$y1][$_] - $target_lab[$x2][$y2][$_])**2 } 0..2);

  } else { # dither mode
    my $xoffmin = ($x1 == 0 || $x2 == 0 ? 0 : -1);
    my $xoffmax = ($x1 == $width - 1 || $x2 == $width - 1 ? 0 : 1);
    my $yoffmin = ($y1 == 0 || $y2 == 0 ? 0 : -1);
    my $yoffmax = ($y1 == $height - 1 || $y2 == $height - 1 ? 0 : 1);

    my (@img1, @img2, @target1, @target2, $points);
    for my $xoff ($xoffmin .. $xoffmax) {
      for my $yoff ($yoffmin .. $yoffmax) {
        $points++;
        for my $chan (0 .. 2) {
          $img1[$chan] += $img_lab[$x1+$xoff][$y1+$yoff][$chan];
          $img2[$chan] += $img_lab[$x2+$xoff][$y2+$yoff][$chan];
          $target1[$chan] += $target_lab[$x1+$xoff][$y1+$yoff][$chan];
          $target2[$chan] += $target_lab[$x2+$xoff][$y2+$yoff][$chan];
        }
      }
    }

    my @img1s = @img1;
    my @img2s = @img2;
    for my $chan (0 .. 2) {
      $img1[$chan] += $img_lab[$x1][$y1][$chan] * ($detail - 1);
      $img2[$chan] += $img_lab[$x2][$y2][$chan] * ($detail - 1);

      $target1[$chan] += $target_lab[$x1][$y1][$chan] * ($detail - 1);
      $target2[$chan] += $target_lab[$x2][$y2][$chan] * ($detail - 1);

      $img1s[$chan] += $img_lab[$x2][$y2][$chan] * $detail - $img_lab[$x1][$y1][$chan];
      $img2s[$chan] += $img_lab[$x1][$y1][$chan] * $detail - $img_lab[$x2][$y2][$chan];
    }

    $dist = (sum map { ($img1[$_] - $target1[$_])**2 } 0..2)
          + (sum map { ($img2[$_] - $target2[$_])**2 } 0..2);

    $dist_swapped = (sum map { ($img1s[$_] - $target1[$_])**2 } 0..2)
                  + (sum map { ($img2s[$_] - $target2[$_])**2 } 0..2);

  }

  if ($dist_swapped < $dist) {
    my @pix1 = $img->GetPixel(x => $x1, y => $y1);
    my @pix2 = $img->GetPixel(x => $x2, y => $y2);
    $img->SetPixel(x => $x1, y => $y1, color => \@pix2);
    $img->SetPixel(x => $x2, y => $y2, color => \@pix1);
    ($img_lab[$x1][$y1], $img_lab[$x2][$y2]) = ($img_lab[$x2][$y2], $img_lab[$x1][$y1]);
    $made ++;
  } else {
    $rejected ++;
  }

  if ($n % 50000 == 0) {
#    print "Made: $made Rejected: $rejected\n";
    $img->Write('png:out.png');
    system("cp", "out.png", sprintf("anim/frame%05d.png", $frame++));
    if ($mode == 1 and $made < $limit1) {
      $mode = 2;
      system("cp", "out.png", "nodither.png");
    } elsif ($mode == 2 and $made < $limit2) {
      last;
    }
    ($made, $rejected) = (0, 0);
  }
}

Resultados

Paleta gótica americana

Pouca diferença aqui com pontilhamento ou não.

Paleta Mona Lisa

O pontilhamento reduz as faixas nas esferas, mas não é especialmente bonito.

Paleta Noite Estrelada

Mona Lisa retém um pouco mais de detalhes com o pontilhamento. Esferas é sobre a mesma situação da última vez.

Paleta de gritos

Noite estrelada sem hesitar é a coisa mais incrível de todas. O pontilhamento torna a foto mais precisa, mas muito menos interessante.

Paleta de esferas

Como diz o aditsu, o verdadeiro teste. Eu acho que passo.

O pontilhamento ajuda imensamente com o gótico americano e a Mona Lisa, misturando tons de cinza e outras cores com os pixels mais intensos para produzir tons de pele semi-precisos em vez de manchas horríveis. O grito é afetado muito menos.

Camaro - Mustang

Imagens de origem da postagem de flawr.

Camaro:

Mustang:

Paleta Camaro

Parece muito bom sem pontilhamento.

Um pontilhamento "apertado" (o mesmo fator de detalhe acima) não muda muito, apenas adiciona um pequeno detalhe nos destaques no capô e no teto.

Um pontilhamento "frouxo" (fator de detalhe reduzido para 6) realmente suaviza a tonalidade, e muito mais detalhes são visíveis no para-brisa, mas os padrões de ditherng são mais óbvios em todos os lugares.

Paleta de Mustang

Partes do carro parecem ótimas, mas os pixels cinzas parecem defeituosos. O que é pior: todos os pixels amarelos mais escuros foram distribuídos pelo corpo vermelho do Camaro, e o algoritmo que não hesita em encontrar nada a ver com os mais leves (movê-los para dentro do carro tornaria a partida pior e movê-los para outro um ponto no fundo não faz diferença líquida), então há um Mustang fantasma no fundo.

O pontilhamento é capaz de espalhar esses pixels amarelos extras para que eles não toquem, espalhando-os mais ou menos uniformemente sobre o plano de fundo no processo. Os destaques e as sombras do carro parecem um pouco melhores.

Novamente, o pontilhamento solto tem a tonalidade mais uniforme, revela mais detalhes nos faróis e no para-brisa. O carro quase parece vermelho novamente. o fundo é mais grosseiro por algum motivo. Não tenho certeza se eu gosto.

Vídeo

( Link HQ )


3
Eu realmente gosto dessa imagem, as imagens muito pontilhadas têm uma sensação maravilhosamente pontilhista . Seurat Mona Lisa alguém?
Boris, a Aranha

2
Seu algoritmo definitivamente faz um ótimo trabalho com a horrível paleta Spheres, bom trabalho!
Corpo de neve

1
@hobbs Fantástico uso da paleta de arco-íris e seus carros são quase perfeitos! Tudo bem se eu usasse algumas de suas imagens em um vídeo do YouTube para mostrar meu script de animação?
9789 Calvin's Hobbies

1
Acho que a única razão pela qual o pontilhamento fornece esse padrão é porque você está usando um bloco de pixels 3x3 com o peso alterado apenas para o centro. Se você ponderou os pixels de acordo com a distância do centro (portanto, os pixels de canto contribuem menos que os 4 adjacentes) e possivelmente se estendeu para um pouco mais de pixels, o pontilhamento deve ser menos perceptível. Já é uma grande melhoria, tais para a paleta arco-íris assim pode valer a pena ver o que mais ele pode fazer ...
Trichoplax

1
@githubphagocyte Passei meio dia tentando coisas assim, mas nada funcionou como eu queria. Uma variante produziu um pontilhamento de aparência aleatória muito bom, mas também me proporcionou uma fase de otimização que nunca terminava. Outras variantes tinham artefatos piores ou pontilhamento muito forte. Minha solução C tem melhor pontilhamento graças à interpolação de splines do ImageMagick. É um spline cúbico, então acho que está usando um bairro 5x5.
Hbbs

79

Pitão

A ideia é simples: todo pixel tem um ponto no espaço RGB 3D. O objetivo é combinar cada um pixel da fonte e um da imagem de destino, de preferência eles devem ser 'próximos' (representam a 'mesma' cor). Como eles podem ser distribuídos de maneiras bem diferentes, não podemos apenas igualar o vizinho mais próximo.

Estratégia

Deixei n Ser um número inteiro (pequeno, 3-255 ou mais). Agora, a nuvem de pixels no espaço RGB é classificada pelo primeiro eixo (R). Agora, esse conjunto de pixels é particionado em n partições. Agora, cada uma das partições é classificada ao longo do segundo eixo (B), que novamente é classificado da mesma maneira particionada. Fazemos isso com as duas figuras e agora temos uma matriz de pontos. Agora podemos apenas combinar os pixels por sua posição na matriz, pois um pixel na mesma posição em cada matriz tem uma posição semelhante em relação a cada pixelcloud no espaço RGB.

Se a distribuição dos pixels no espaço RGB das duas imagens for semelhante (significa apenas deslocado e / ou esticado ao longo do eixo 3), o resultado será bastante previsível. Se as distribuições parecerem completamente diferentes, esse algoritmo não produzirá bons resultados (como visto no último exemplo), mas esse também é um dos casos mais difíceis que eu acho. O que isso não faz é usar efeitos de interação de pixels vizinhos na percepção.

Código

Disclaimer: Eu sou um novato absoluto em python.

from PIL import Image

n = 5 #number of partitions per channel.

src_index = 3 #index of source image
dst_index = 2 #index of destination image

images =  ["img0.bmp","img1.bmp","img2.bmp","img3.bmp"];
src_handle = Image.open(images[src_index])
dst_handle = Image.open(images[dst_index])
src = src_handle.load()
dst = dst_handle.load()
assert src_handle.size[0]*src_handle.size[1] == dst_handle.size[0]*dst_handle.size[1],"images must be same size"

def makePixelList(img):
    l = []
    for x in range(img.size[0]):
        for y in range(img.size[1]):
            l.append((x,y))
    return l

lsrc = makePixelList(src_handle)
ldst = makePixelList(dst_handle)

def sortAndDivide(coordlist,pixelimage,channel): #core
    global src,dst,n
    retlist = []
    #sort
    coordlist.sort(key=lambda t: pixelimage[t][channel])
    #divide
    partitionLength = int(len(coordlist)/n)
    if partitionLength <= 0:
        partitionLength = 1
    if channel < 2:
        for i in range(0,len(coordlist),partitionLength):
            retlist += sortAndDivide(coordlist[i:i+partitionLength],pixelimage,channel+1)
    else:
        retlist += coordlist
    return retlist

print(src[lsrc[0]])

lsrc = sortAndDivide(lsrc,src,0)
ldst = sortAndDivide(ldst,dst,0)

for i in range(len(ldst)):
    dst[ldst[i]] = src[lsrc[i]]

dst_handle.save("exchange"+str(src_index)+str(dst_index)+".png")

Resultado

Eu acho que ficou ruim considerando a solução simples. Obviamente, é possível obter melhores resultados quando se brinca com o parâmetro, ou primeiro transformando as cores em outro espaço de cores, ou até otimizando o particionamento.

comparação dos meus resultados

Galeria completa aqui: https://imgur.com/a/hzaAm#6

Detalhado para River

monalisa> rio

monalisa> rio

pessoas> rio

pessoas> rio

bolas> rio

bolas> rio

noite estrelada> rio

noturno> rio

o choro> rio

thecry> rio

bolas> MonaLisa, variando n = 2,4,6, ..., 20

Essa foi a tarefa mais desafiadora que eu acho, longe de belas imagens, aqui um gif (teve que ser reduzido para 256 cores) com os diferentes valores de parâmetros n = 2,4,6, ..., 20. Para mim, foi surpreendente que valores muito baixos produzissem melhores imagens (ao olhar para o rosto da sra. Lisa): bolas> monalisa

Desculpe, eu não consigo parar

Qual você gosta mais? Chevy Camaro ou Ford Mustang? Talvez essa técnica possa ser melhorada e usada para colorir imagens em preto e branco. Agora aqui: primeiro eu cortei os carros mais ou menos do fundo pintando-os de branco (em tinta, não muito profissional ...) e depois usei o programa python em cada direção.

Originais

original original

Recoloured

Acho que existem artefatos porque a área de um carro era um pouco maior que a outra e porque minhas habilidades artísticas são muito ruins =) manipulado insira a descrição da imagem aqui


5
Uau, eu realmente amo o rio da Noite Estrelada, e como o Grito faz parecer um rio de fogo.
9789 Calvin's Hobbies

@ Calvin'sHobbies uau sim! Eles quase parecem atraídos, eu nem os olhei de perto desde que eu estava ocupado carregando as novas imagens = P Mas obrigado por este grande desafio!
flawr

3
Eu amo as transformações do carro. Isso pode se tornar um tipo de transformação de edição de imagem, realmente!
tomsmeding

@tomsmeding Obrigado, eu já pensei em usar a técnica para a colorização de imagens em preto e branco, mas até agora com sucesso limitado. Mas talvez precisamos de mais algumas idéias para conseguir este feito =)
flawr

@flawr Estaria tudo bem se eu usasse algumas de suas imagens em um vídeo do YouTube para mostrar meu script de animação?
9789 Calvin's Hobbies

48

Python - Uma solução teoricamente ideal

Eu digo teoricamente ideal porque a solução realmente ótima não é viável de calcular. Começo descrevendo a solução teórica e depois explico como a aprimorei para torná-la viável computacionalmente no espaço e no tempo.

Considero a solução mais ideal como a que produz o menor erro total em todos os pixels entre as imagens antigas e as novas. O erro entre dois pixels é definido como a distância euclidiana entre os pontos no espaço 3D em que cada valor de cor (R, G, B) é uma coordenada. Na prática, devido à maneira como os humanos veem as coisas, a solução ideal pode muito bem não ser a solução mais bonita . No entanto, parece funcionar bastante bem em todos os casos.

Para calcular o mapeamento, considerei isso um problema de correspondência bipartida com peso mínimo . Em outras palavras, existem dois conjuntos de nós: os pixels originais e os pixels da paleta. Uma aresta é criada entre cada pixel nos dois conjuntos (mas nenhuma aresta é criada dentro de um conjunto). O custo, ou peso, de uma aresta é a distância euclidiana entre os dois pixels, conforme descrito acima. Quanto mais próximas as cores estiverem visualmente, menor será o custo entre os pixels.

Exemplo de correspondência bipartida

Isso cria uma matriz de custo do tamanho N 2 . Para essas imagens em que N = 123520, são necessários aproximadamente 40 GB de memória para representar os custos como números inteiros e a metade disso como números curtos. De qualquer maneira, eu não tinha memória suficiente na minha máquina para fazer uma tentativa. Outra questão é que o algoritmo húngaro , ou o algoritmo Jonker-Volgenant , que pode ser usado para resolver esse problema, roda em N 3 tempo . Embora definitivamente computável, a geração de uma solução por imagem provavelmente levaria horas ou dias.

Para contornar esse problema, ordeno aleatoriamente as duas listas de pixels, divido as listas em pedaços C, execute uma implementação C ++ do algoritmo Jonker-Volgenant em cada par de sub-listas e, em seguida, ingresso nas listas novamente para criar o mapeamento final. Portanto, o código abaixo permitiria encontrar a solução realmente ideal, desde que eles definissem o tamanho do fragmento C para 1 (sem fragmento) e tivessem memória suficiente. Para essas imagens, defino C como 16, para que N se torne 7720, levando apenas alguns minutos por imagem.

Uma maneira simples de pensar por que isso funciona é que a classificação aleatória da lista de pixels e a captura de um subconjunto é como amostrar a imagem. Portanto, ao definir C = 16, é como tirar 16 amostras aleatórias distintas de tamanho N / C do original e da paleta. Concedido, provavelmente existem maneiras melhores de dividir as listas, mas uma abordagem aleatória fornece resultados decentes.

import subprocess
import multiprocessing as mp
import sys
import os
import sge
from random import shuffle
from PIL import Image
import numpy as np
import LAPJV
import pdb

def getError(p1, p2):
    return (p1[0]-p2[0])**2 + (p1[1]-p2[1])**2 + (p1[2]-p2[2])**2

def getCostMatrix(pallete_list, source_list):
    num_pixels = len(pallete_list)
    matrix = np.zeros((num_pixels, num_pixels))

    for i in range(num_pixels):
        for j in range(num_pixels):
            matrix[i][j] = getError(pallete_list[i], source_list[j])

    return matrix

def chunks(l, n):
    if n < 1:
        n = 1
    return [l[i:i + n] for i in range(0, len(l), n)]

def imageToColorList(img_file):
    i = Image.open(img_file)

    pixels = i.load()
    width, height = i.size

    all_pixels = []
    for x in range(width):
        for y in range(height):
            pixel = pixels[x, y]
            all_pixels.append(pixel)

    return all_pixels

def colorListToImage(color_list, old_img_file, new_img_file, mapping):
    i = Image.open(old_img_file)

    pixels = i.load()
    width, height = i.size
    idx = 0

    for x in range(width):
        for y in range(height):
            pixels[x, y] = color_list[mapping[idx]]
            idx += 1

    i.save(new_img_file)

def getMapping(pallete_list, source_list):
    matrix = getCostMatrix(source_list, pallete_list)
    result = LAPJV.lap(matrix)[1]
    ret = []
    for i in range(len(pallete_list)):
        ret.append(result[i])
    return ret

def randomizeList(l):
    rdm_l = list(l)
    shuffle(rdm_l)
    return rdm_l

def getPartialMapping(zipped_chunk):
    pallete_chunk = zipped_chunk[0]
    source_chunk = zipped_chunk[1]
    subl_pallete = map(lambda v: v[1], pallete_chunk)
    subl_source = map(lambda v: v[1], source_chunk)
    mapping = getMapping(subl_pallete, subl_source)
    return mapping

def getMappingWithPartitions(pallete_list, source_list, C = 1):
    rdm_pallete = randomizeList(enumerate(pallete_list))
    rdm_source = randomizeList(enumerate(source_list))
    num_pixels = len(rdm_pallete)
    real_mapping = [0] * num_pixels

    chunk_size = int(num_pixels / C)

    chunked_rdm_pallete = chunks(rdm_pallete, chunk_size)
    chunked_rdm_source = chunks(rdm_source, chunk_size)
    zipped_chunks = zip(chunked_rdm_pallete, chunked_rdm_source)

    pool = mp.Pool(2)
    mappings = pool.map(getPartialMapping, zipped_chunks)

    for mapping, zipped_chunk in zip(mappings, zipped_chunks):
        pallete_chunk = zipped_chunk[0]
        source_chunk = zipped_chunk[1]
        for idx1,idx2 in enumerate(mapping):
            src_px = source_chunk[idx1]
            pal_px = pallete_chunk[idx2]
            real_mapping[src_px[0]] = pal_px[0]

    return real_mapping

def run(pallete_name, source_name, output_name):
    print("Getting Colors...")
    pallete_list = imageToColorList(pallete_name)
    source_list = imageToColorList(source_name)

    print("Creating Mapping...")
    mapping = getMappingWithPartitions(pallete_list, source_list, C = 16)

    print("Generating Image...");
    colorListToImage(pallete_list, source_name, output_name, mapping)

if __name__ == '__main__':
    pallete_name = sys.argv[1]
    source_name = sys.argv[2]
    output_name = sys.argv[3]
    run(pallete_name, source_name, output_name)

Resultados:

Como na solução do aditsu, essas imagens foram geradas usando exatamente os mesmos parâmetros. O único parâmetro aqui é C, que deve ser definido o mais baixo possível. Para mim, C = 16 foi um bom equilíbrio entre velocidade e qualidade.

Todas as imagens: http://imgur.com/a/RCZiX#0

Paleta gótica americana

mona-gótico grito-gótico

Paleta Mona Lisa

gothic-mona grito-mona

Paleta Noite Estrelada

mona-noite noite do rio

Paleta de gritos

grito gótico grito mona

Paleta River

esferas góticas mona-esferas

Paleta de esferas

esferas góticas mona-esferas


4
Eu realmente gosto (Grito -> Noite estrelada) e (Esferas -> Noite estrelada). (Esferas -> Mona Lisa) também não é tão ruim, mas eu gostaria de ver mais hesitações.
John Dvorak

Lol, eu estava pensando o mesmo sobre a correspondência grafo bipartido, mas abandonded a idéia porque o N ^ 3 ..
Robau

Esse algoritmo "quase determinístico" supera todos os determinísticos da IMO e se destaca dos bons randomizados. Eu gosto disso.
Hbbs

1
Não concordo com a sua noção de uma solução ideal. Por quê? O pontilhamento pode melhorar a qualidade perceptiva (para humanos) e, no entanto, gerar uma pontuação mais baixa usando sua definição. Também usar RGB sobre algo como CIELUV é um erro.
Thomas Eding

39

Pitão

Edit: Acabei de perceber que você pode realmente aprimorar a fonte com o ImageFilter para tornar os resultados mais bem definidos.

Arco-íris -> Mona Lisa (fonte afiada de Mona Lisa, apenas Luminância)

insira a descrição da imagem aqui

Arco-íris -> Mona Lisa (fonte não afiada, ponderada com Y = 10, I = 10, Q = 0)

insira a descrição da imagem aqui

Mona Lisa -> Gótico Americano (fonte não afiada, apenas Luminância)

insira a descrição da imagem aqui

Mona Lisa -> Gótico Americano (fonte não afiada, ponderada com Y = 1, I = 10, Q = 1)

insira a descrição da imagem aqui

Rio -> Arco-íris (fonte não afiada, apenas luminância)

insira a descrição da imagem aqui

Basicamente, obtém todos os pixels das duas imagens em duas listas.

Classifique-os com a luminância como chave. Y em YIQ representa a luminância.

Em seguida, para cada pixel na fonte (que está em ordem crescente de luminância), obtenha o valor RGB a partir de pixels do mesmo índice na lista de paletes.

import Image, ImageFilter, colorsys

def getPixels(image):
    width, height = image.size
    pixels = []
    for x in range(width):
        for y in range(height):
            pixels.append([(x,y), image.getpixel((x,y))])
    return pixels

def yiq(pixel):
    # y is the luminance
    y,i,q = colorsys.rgb_to_yiq(pixel[1][0], pixel[1][6], pixel[1][7])
    # Change the weights accordingly to get different results
    return 10*y + 0*i + 0*q

# Open the images
source  = Image.open('ml.jpg')
pallete = Image.open('rainbow.png')

# Sharpen the source... It won't affect the palette anyway =D
source = source.filter(ImageFilter.SHARPEN)

# Sort the two lists by luminance
sourcePixels  = sorted(getPixels(source),  key=yiq)
palletePixels = sorted(getPixels(pallete), key=yiq)

copy = Image.new('RGB', source.size)

# Iterate through all the coordinates of source
# And set the new color
index = 0
for sourcePixel in sourcePixels:
    copy.putpixel(sourcePixel[0], palletePixels[index][8])
    index += 1

# Save the result
copy.save('copy.png')

Para acompanhar a tendência das animações ...

Pixels no grito sendo classificados rapidamente na noite estrelada e vice-versa

insira a descrição da imagem aqui insira a descrição da imagem aqui


2
Essa ideia simples funciona muito bem. Gostaria de saber se ele pode ser estendido e usar luminância ponderada, saturação e matiz. (Por exemplo, 10 * L + S + H) para obter uma melhor correspondência de cores na mesma área.
Moogie

1
@bitpwnr Suas imagens não estão passando no meu script, mas isso é quase certo porque você está usando os jpegs ligeiramente diferentes que eu criei inicialmente, então não é grande coisa. No entanto, só consegui executar seu código após substituir [6], [7] e [8] por [1], [2] e [1]. Estou recebendo as mesmas imagens, mas esse é um erro muito único: P
Calvin's Hobbies

Suas imagens são muito claras, mas meio que desaturated: p
aditsu

@ Calvin'sHobbies Opps, corrigiu os erros de digitação.
Vetorizado

@bitpwner Seria bom se eu usasse algumas de suas imagens em um vídeo do YouTube para mostrar meu script de animação?
9789 Calvin's Hobbies

39

C # Winform - Visual Studio 2010

Editar pontilhamento adicionado.

Essa é a minha versão do algoritmo de troca aleatória - sabor @hobbs. Ainda sinto que algum tipo de pontilhamento não aleatório pode fazer melhor ...

Elaboração de cores no espaço Y-Cb-Cr (como na compressão jpeg)

Elaboração em duas fases:

  1. Cópia de pixel da fonte em ordem de luminância. Isso já dá uma boa imagem, mas desaturada - quase na escala de cinza - em quase 0 tempo
  2. Troca aleatória repetida de pixels. A troca é feita se isso der um delta melhor (em relação à fonte) na célula 3x3 que contém o pixel. Portanto, é um efeito pontilhado. O delta é calculado no espaço Y-Cr-Cb sem ponderação dos diferentes componentes.

Esse é essencialmente o mesmo método usado pelo @hobbs, sem a primeira troca aleatória sem pontilhamento. Meus tempos são mais curtos (o idioma conta?) E acho que minhas imagens são melhores (provavelmente o espaço de cores usado é mais preciso).

Uso do programa: coloque imagens .png na pasta c: \ temp, marque o elemento na lista para escolher a imagem da paleta, selecione o elemento na lista para escolher a imagem de origem (não tão fácil de usar). Clique no botão Iniciar para iniciar a elaboração, o salvamento é automático (mesmo quando você preferir não tomar cuidado).

Tempo de elaboração inferior a 90 segundos.

Resultados atualizados

Paleta: Gótico Americano

Monna Lisa arco Iris Rio Grito Noite estrelada

Paleta: Monna Lisa

gótico americano arco Iris Rio Grito Noite estrelada

Paleta: Arco-íris

gótico americano Monna Lisa Rio Grito Noite estrelada

Paleta: Rio

gótico americano Monna Lisa arco Iris Grito Noite estrelada

Paleta: Grito

gótico americano Monna Lisa arco Iris Rio Noite estrelada

Paleta: Noite Estrelada

gótico americano Monna Lisa arco Iris Rio Grito

Form1.cs

using System;
using System.Collections.Generic;
using System.ComponentModel;
using System.Data;
using System.Drawing;
using System.Linq;
using System.Text;
using System.Windows.Forms;
using System.Drawing.Imaging;
using System.IO;

namespace Palette
{
    public struct YRB
    {
        public int y, cb, cr;

        public YRB(int r, int g, int b)
        {
            y = (int)(0.299 * r + 0.587 * g + 0.114 * b);
            cb = (int)(128 - 0.168736 * r - 0.331264 * g + 0.5 * b);
            cr = (int)(128 + 0.5 * r - 0.418688 * g - 0.081312 * b);
        }
    }

    public struct Pixel
    {
        private const int ARGBAlphaShift = 24;
        private const int ARGBRedShift = 16;
        private const int ARGBGreenShift = 8;
        private const int ARGBBlueShift = 0;

        public int px, py;
        private uint _color;
        public YRB yrb;

        public Pixel(uint col, int px = 0, int py = 0)
        {
            this.px = px;
            this.py = py;
            this._color = col;
            yrb = new YRB((int)(col >> ARGBRedShift) & 255, (int)(col >> ARGBGreenShift) & 255, (int)(col >> ARGBBlueShift) & 255); 
        }

        public uint color
        {
            get { 
                return _color; 
            }
            set {
                _color = color;
                yrb = new YRB((int)(color >> ARGBRedShift) & 255, (int)(color >> ARGBGreenShift) & 255, (int)(color >> ARGBBlueShift) & 255);
            }
        }

        public int y
        {
            get { return yrb.y; }
        }
        public int cr
        {
            get { return yrb.cr; }
        }
        public int cb
        {
            get { return yrb.cb; }
        }
    }

    public partial class Form1 : Form
    {
        public Form1()
        {
            InitializeComponent();
        }

        private void Form1_Load(object sender, EventArgs e)
        {
            DirectoryInfo di = new System.IO.DirectoryInfo(@"c:\temp\");
            foreach (FileInfo file in di.GetFiles("*.png"))
            {
                ListViewItem item = new ListViewItem(file.Name);
                item.SubItems.Add(file.FullName);
                lvFiles.Items.Add(item);
            }
        }

        private void lvFiles_ItemSelectionChanged(object sender, ListViewItemSelectionChangedEventArgs e)
        {
            if (e.IsSelected)
            {
                string file = e.Item.SubItems[1].Text;
                GetImagePB(pbSource, file);
                pbSource.Tag = file; 
                DupImage(pbSource, pbOutput);

                this.Width = pbOutput.Width + pbOutput.Left + 20;
                this.Height = Math.Max(pbOutput.Height, pbPalette.Height)+lvFiles.Height*2;   
            }
        }

        private void lvFiles_ItemCheck(object sender, ItemCheckEventArgs e)
        {
            foreach (ListViewItem item in lvFiles.CheckedItems)
            {
                if (item.Index != e.Index) item.Checked = false;
            }
            string file = lvFiles.Items[e.Index].SubItems[1].Text;
            GetImagePB(pbPalette, file);
            pbPalette.Tag = lvFiles.Items[e.Index].SubItems[0].Text; 

            this.Width = pbOutput.Width + pbOutput.Left + 20;
            this.Height = Math.Max(pbOutput.Height, pbPalette.Height) + lvFiles.Height * 2;   
        }

        Pixel[] Palette;
        Pixel[] Source;

        private void BtnStart_Click(object sender, EventArgs e)
        {
            lvFiles.Enabled = false;
            btnStart.Visible = false;
            progressBar.Visible = true; 
            DupImage(pbSource, pbOutput);

            Work(pbSource.Image as Bitmap, pbPalette.Image as Bitmap, pbOutput.Image as Bitmap);

            string newfile = (string)pbSource.Tag +"-"+ (string)pbPalette.Tag;
            pbOutput.Image.Save(newfile, ImageFormat.Png);   

            lvFiles.Enabled = true;
            btnStart.Visible = true;
            progressBar.Visible = false;
        }

        private void Work(Bitmap srcb, Bitmap palb, Bitmap outb)
        {
            GetData(srcb, out Source);
            GetData(palb, out Palette);

            FastBitmap fout = new FastBitmap(outb);
            FastBitmap fsrc = new FastBitmap(srcb);
            int pm = Source.Length;
            int w = outb.Width;
            int h = outb.Height;
            progressBar.Maximum = pm;

            fout.LockImage();
            for (int p = 0; p < pm; p++)
            {
                fout.SetPixel(Source[p].px, Source[p].py, Palette[p].color);
            }
            fout.UnlockImage();

            pbOutput.Refresh();

            var rnd = new Random();
            int totsw = 0;
            progressBar.Maximum = 200;
            for (int i = 0; i < 200; i++)
            {
                int nsw = 0;
                progressBar.Value = i;
                fout.LockImage();
                fsrc.LockImage();
                for (int j = 0; j < 200000; j++)
                {
                    nsw += CheckSwap(fsrc, fout, 1 + rnd.Next(w - 2), 1 + rnd.Next(h - 2), 1 + rnd.Next(w - 2), 1 + rnd.Next(h - 2));
                }
                totsw += nsw;
                lnCurSwap.Text = nsw.ToString();
                lnTotSwap.Text = totsw.ToString();
                fout.UnlockImage();
                fsrc.UnlockImage();
                pbOutput.Refresh();
                Application.DoEvents();
                if (nsw == 0)
                {
                    break;
                }
            }            
        }

        int CheckSwap(FastBitmap fsrc, FastBitmap fout, int x1, int y1, int x2, int y2)
        {
            const int fmax = 3;
            YRB ov1 = new YRB();
            YRB sv1 = new YRB();
            YRB ov2 = new YRB();
            YRB sv2 = new YRB();

            int f;
            for (int dx = -1; dx <= 1; dx++)
            {
                for (int dy = -1; dy <= 1; dy++)
                {
                    f = (fmax - Math.Abs(dx) - Math.Abs(dy));
                    {
                        Pixel o1 = new Pixel(fout.GetPixel(x1 + dx, y1 + dy));
                        ov1.y += o1.y * f;
                        ov1.cb += o1.cr * f;
                        ov1.cr += o1.cb * f;

                        Pixel s1 = new Pixel(fsrc.GetPixel(x1 + dx, y1 + dy));
                        sv1.y += s1.y * f;
                        sv1.cb += s1.cr * f;
                        sv1.cr += s1.cb * f;

                        Pixel o2 = new Pixel(fout.GetPixel(x2 + dx, y2 + dy));
                        ov2.y += o2.y * f;
                        ov2.cb += o2.cr * f;
                        ov2.cr += o2.cb * f;

                        Pixel s2 = new Pixel(fsrc.GetPixel(x2 + dx, y2 + dy));
                        sv2.y += s2.y * f;
                        sv2.cb += s2.cr * f;
                        sv2.cr += s2.cb * f;
                    }
                }
            }
            YRB ox1 = ov1;
            YRB ox2 = ov2;
            Pixel oc1 = new Pixel(fout.GetPixel(x1, y1));
            Pixel oc2 = new Pixel(fout.GetPixel(x2, y2));
            ox1.y += fmax * oc2.y - fmax * oc1.y;
            ox1.cb += fmax * oc2.cr - fmax * oc1.cr;
            ox1.cr += fmax * oc2.cb - fmax * oc1.cb;
            ox2.y += fmax * oc1.y - fmax * oc2.y;
            ox2.cb += fmax  * oc1.cr - fmax * oc2.cr;
            ox2.cr += fmax * oc1.cb - fmax * oc2.cb;

            int curd = Delta(ov1, sv1, 1) + Delta(ov2, sv2, 1);
            int newd = Delta(ox1, sv1, 1) + Delta(ox2, sv2, 1);
            if (newd < curd)
            {
                fout.SetPixel(x1, y1, oc2.color);
                fout.SetPixel(x2, y2, oc1.color);
                return 1;
            }
            return 0;
        }

        int Delta(YRB p1, YRB p2, int sf)
        {
            int dy = (p1.y - p2.y);
            int dr = (p1.cr - p2.cr);
            int db = (p1.cb - p2.cb);

            return dy * dy * sf + dr * dr + db * db;
        }

        Bitmap GetData(Bitmap bmp, out Pixel[] Output)
        {
            FastBitmap fb = new FastBitmap(bmp);
            BitmapData bmpData = fb.LockImage(); 

            Output = new Pixel[bmp.Width * bmp.Height];

            int p = 0;
            for (int y = 0; y < bmp.Height; y++)
            {
                uint col = fb.GetPixel(0, y);
                Output[p++] = new Pixel(col, 0, y);

                for (int x = 1; x < bmp.Width; x++)
                {
                    col = fb.GetNextPixel();
                    Output[p++] = new Pixel(col, x, y);
                }
            }
            fb.UnlockImage(); // Unlock the bits.

            Array.Sort(Output, (a, b) => a.y - b.y);

            return bmp;
        }

        void DupImage(PictureBox s, PictureBox d)
        {
            if (d.Image != null)
                d.Image.Dispose();
            d.Image = new Bitmap(s.Image.Width, s.Image.Height);  
        }

        void GetImagePB(PictureBox pb, string file)
        {
            Bitmap bms = new Bitmap(file, false);
            Bitmap bmp = bms.Clone(new Rectangle(0, 0, bms.Width, bms.Height), PixelFormat.Format32bppArgb);
            bms.Dispose(); 
            if (pb.Image != null)
                pb.Image.Dispose();
            pb.Image = bmp;
        }
    }

    //Adapted from Visual C# Kicks - http://www.vcskicks.com/
    unsafe public class FastBitmap
    {
        private Bitmap workingBitmap = null;
        private int width = 0;
        private BitmapData bitmapData = null;
        private Byte* pBase = null;

        public FastBitmap(Bitmap inputBitmap)
        {
            workingBitmap = inputBitmap;
        }

        public BitmapData LockImage()
        {
            Rectangle bounds = new Rectangle(Point.Empty, workingBitmap.Size);

            width = (int)(bounds.Width * 4 + 3) & ~3;

            //Lock Image
            bitmapData = workingBitmap.LockBits(bounds, ImageLockMode.ReadWrite, PixelFormat.Format32bppArgb);
            pBase = (Byte*)bitmapData.Scan0.ToPointer();
            return bitmapData;
        }

        private uint* pixelData = null;

        public uint GetPixel(int x, int y)
        {
            pixelData = (uint*)(pBase + y * width + x * 4);
            return *pixelData;
        }

        public uint GetNextPixel()
        {
            return *++pixelData;
        }

        public void GetPixelArray(int x, int y, uint[] Values, int offset, int count)
        {
            pixelData = (uint*)(pBase + y * width + x * 4);
            while (count-- > 0)
            {
                Values[offset++] = *pixelData++;
            }
        }

        public void SetPixel(int x, int y, uint color)
        {
            pixelData = (uint*)(pBase + y * width + x * 4);
            *pixelData = color;
        }

        public void SetNextPixel(uint color)
        {
            *++pixelData = color;
        }

        public void UnlockImage()
        {
            workingBitmap.UnlockBits(bitmapData);
            bitmapData = null;
            pBase = null;
        }
    }

}

Form1.Designer.cs

namespace Palette
{
    partial class Form1
    {
        /// <summary>
        /// Variabile di progettazione necessaria.
        /// </summary>
        private System.ComponentModel.IContainer components = null;

        /// <summary>
        /// Liberare le risorse in uso.
        /// </summary>
        /// <param name="disposing">ha valore true se le risorse gestite devono essere eliminate, false in caso contrario.</param>
        protected override void Dispose(bool disposing)
        {
            if (disposing && (components != null))
            {
                components.Dispose();
            }
            base.Dispose(disposing);
        }

        #region Codice generato da Progettazione Windows Form

        /// <summary>
        /// Metodo necessario per il supporto della finestra di progettazione. Non modificare
        /// il contenuto del metodo con l'editor di codice.
        /// </summary>
        private void InitializeComponent()
        {
            this.components = new System.ComponentModel.Container();
            this.panel = new System.Windows.Forms.FlowLayoutPanel();
            this.pbSource = new System.Windows.Forms.PictureBox();
            this.pbPalette = new System.Windows.Forms.PictureBox();
            this.pbOutput = new System.Windows.Forms.PictureBox();
            this.btnStart = new System.Windows.Forms.Button();
            this.progressBar = new System.Windows.Forms.ProgressBar();
            this.imageList1 = new System.Windows.Forms.ImageList(this.components);
            this.lvFiles = new System.Windows.Forms.ListView();
            this.lnTotSwap = new System.Windows.Forms.Label();
            this.lnCurSwap = new System.Windows.Forms.Label();
            this.panel.SuspendLayout();
            ((System.ComponentModel.ISupportInitialize)(this.pbSource)).BeginInit();
            ((System.ComponentModel.ISupportInitialize)(this.pbPalette)).BeginInit();
            ((System.ComponentModel.ISupportInitialize)(this.pbOutput)).BeginInit();
            this.SuspendLayout();
            // 
            // panel
            // 
            this.panel.AutoScroll = true;
            this.panel.AutoSize = true;
            this.panel.Controls.Add(this.pbSource);
            this.panel.Controls.Add(this.pbPalette);
            this.panel.Controls.Add(this.pbOutput);
            this.panel.Dock = System.Windows.Forms.DockStyle.Top;
            this.panel.Location = new System.Drawing.Point(0, 0);
            this.panel.Name = "panel";
            this.panel.Size = new System.Drawing.Size(748, 266);
            this.panel.TabIndex = 3;
            this.panel.WrapContents = false;
            // 
            // pbSource
            // 
            this.pbSource.BorderStyle = System.Windows.Forms.BorderStyle.FixedSingle;
            this.pbSource.Location = new System.Drawing.Point(3, 3);
            this.pbSource.Name = "pbSource";
            this.pbSource.Size = new System.Drawing.Size(157, 260);
            this.pbSource.SizeMode = System.Windows.Forms.PictureBoxSizeMode.AutoSize;
            this.pbSource.TabIndex = 1;
            this.pbSource.TabStop = false;
            // 
            // pbPalette
            // 
            this.pbPalette.BorderStyle = System.Windows.Forms.BorderStyle.FixedSingle;
            this.pbPalette.Location = new System.Drawing.Point(166, 3);
            this.pbPalette.Name = "pbPalette";
            this.pbPalette.Size = new System.Drawing.Size(172, 260);
            this.pbPalette.SizeMode = System.Windows.Forms.PictureBoxSizeMode.AutoSize;
            this.pbPalette.TabIndex = 3;
            this.pbPalette.TabStop = false;
            // 
            // pbOutput
            // 
            this.pbOutput.BorderStyle = System.Windows.Forms.BorderStyle.FixedSingle;
            this.pbOutput.Location = new System.Drawing.Point(344, 3);
            this.pbOutput.Name = "pbOutput";
            this.pbOutput.Size = new System.Drawing.Size(172, 260);
            this.pbOutput.SizeMode = System.Windows.Forms.PictureBoxSizeMode.AutoSize;
            this.pbOutput.TabIndex = 4;
            this.pbOutput.TabStop = false;
            // 
            // btnStart
            // 
            this.btnStart.Anchor = ((System.Windows.Forms.AnchorStyles)((System.Windows.Forms.AnchorStyles.Bottom | System.Windows.Forms.AnchorStyles.Right)));
            this.btnStart.Location = new System.Drawing.Point(669, 417);
            this.btnStart.Name = "btnStart";
            this.btnStart.Size = new System.Drawing.Size(79, 42);
            this.btnStart.TabIndex = 4;
            this.btnStart.Text = "Start";
            this.btnStart.UseVisualStyleBackColor = true;
            this.btnStart.Click += new System.EventHandler(this.BtnStart_Click);
            // 
            // progressBar
            // 
            this.progressBar.Dock = System.Windows.Forms.DockStyle.Bottom;
            this.progressBar.Location = new System.Drawing.Point(0, 465);
            this.progressBar.Name = "progressBar";
            this.progressBar.Size = new System.Drawing.Size(748, 16);
            this.progressBar.TabIndex = 5;
            // 
            // imageList1
            // 
            this.imageList1.ColorDepth = System.Windows.Forms.ColorDepth.Depth8Bit;
            this.imageList1.ImageSize = new System.Drawing.Size(16, 16);
            this.imageList1.TransparentColor = System.Drawing.Color.Transparent;
            // 
            // lvFiles
            // 
            this.lvFiles.Anchor = ((System.Windows.Forms.AnchorStyles)(((System.Windows.Forms.AnchorStyles.Bottom | System.Windows.Forms.AnchorStyles.Left) 
            | System.Windows.Forms.AnchorStyles.Right)));
            this.lvFiles.CheckBoxes = true;
            this.lvFiles.HideSelection = false;
            this.lvFiles.Location = new System.Drawing.Point(12, 362);
            this.lvFiles.MultiSelect = false;
            this.lvFiles.Name = "lvFiles";
            this.lvFiles.Size = new System.Drawing.Size(651, 97);
            this.lvFiles.Sorting = System.Windows.Forms.SortOrder.Ascending;
            this.lvFiles.TabIndex = 7;
            this.lvFiles.UseCompatibleStateImageBehavior = false;
            this.lvFiles.View = System.Windows.Forms.View.List;
            this.lvFiles.ItemCheck += new System.Windows.Forms.ItemCheckEventHandler(this.lvFiles_ItemCheck);
            this.lvFiles.ItemSelectionChanged += new System.Windows.Forms.ListViewItemSelectionChangedEventHandler(this.lvFiles_ItemSelectionChanged);
            // 
            // lnTotSwap
            // 
            this.lnTotSwap.Anchor = ((System.Windows.Forms.AnchorStyles)((System.Windows.Forms.AnchorStyles.Bottom | System.Windows.Forms.AnchorStyles.Right)));
            this.lnTotSwap.Location = new System.Drawing.Point(669, 362);
            this.lnTotSwap.Name = "lnTotSwap";
            this.lnTotSwap.Size = new System.Drawing.Size(58, 14);
            this.lnTotSwap.TabIndex = 8;
            this.lnTotSwap.Text = "label1";
            // 
            // lnCurSwap
            // 
            this.lnCurSwap.Anchor = ((System.Windows.Forms.AnchorStyles)((System.Windows.Forms.AnchorStyles.Bottom | System.Windows.Forms.AnchorStyles.Right)));
            this.lnCurSwap.Location = new System.Drawing.Point(669, 385);
            this.lnCurSwap.Name = "lnCurSwap";
            this.lnCurSwap.Size = new System.Drawing.Size(58, 14);
            this.lnCurSwap.TabIndex = 9;
            this.lnCurSwap.Text = "label1";
            // 
            // Form1
            // 
            this.AutoScaleDimensions = new System.Drawing.SizeF(6F, 13F);
            this.AutoScaleMode = System.Windows.Forms.AutoScaleMode.Font;
            this.BackColor = System.Drawing.SystemColors.ControlDark;
            this.ClientSize = new System.Drawing.Size(748, 481);
            this.Controls.Add(this.lnCurSwap);
            this.Controls.Add(this.lnTotSwap);
            this.Controls.Add(this.lvFiles);
            this.Controls.Add(this.progressBar);
            this.Controls.Add(this.btnStart);
            this.Controls.Add(this.panel);
            this.Name = "Form1";
            this.Text = "Form1";
            this.Load += new System.EventHandler(this.Form1_Load);
            this.panel.ResumeLayout(false);
            this.panel.PerformLayout();
            ((System.ComponentModel.ISupportInitialize)(this.pbSource)).EndInit();
            ((System.ComponentModel.ISupportInitialize)(this.pbPalette)).EndInit();
            ((System.ComponentModel.ISupportInitialize)(this.pbOutput)).EndInit();
            this.ResumeLayout(false);
            this.PerformLayout();

        }

        #endregion

        private System.Windows.Forms.FlowLayoutPanel panel;
        private System.Windows.Forms.PictureBox pbSource;
        private System.Windows.Forms.PictureBox pbPalette;
        private System.Windows.Forms.PictureBox pbOutput;
        private System.Windows.Forms.Button btnStart;
        private System.Windows.Forms.ProgressBar progressBar;
        private System.Windows.Forms.ImageList imageList1;
        private System.Windows.Forms.ListView lvFiles;
        private System.Windows.Forms.Label lnTotSwap;
        private System.Windows.Forms.Label lnCurSwap;
    }
}

Program.cs

using System;
using System.Collections.Generic;
using System.Linq;
using System.Windows.Forms;

namespace Palette
{
    static class Program
    {
        /// <summary>
        /// Punto di ingresso principale dell'applicazione.
        /// </summary>
        [STAThread]
        static void Main()
        {
            Application.EnableVisualStyles();
            Application.SetCompatibleTextRenderingDefault(false);
            Application.Run(new Form1());
        }
    }
}

Marque 'Código não seguro' na propriedade do projeto para compilar.


4
O IMO produz os melhores resultados #
figgycity50

9
Isso é absolutamente incrível com essa paleta de arco-íris horrível.
Michael B

1
Incrível, vencedor!
Jjrv 17/07

25

JS

Basta executar em dois URLs de imagem.

Como um pacote JS, você pode executá-lo sozinho no navegador. São fornecidos violinos que brincam com configurações diferentes. Lembre-se de que este violino: http://jsfiddle.net/eithe/J7jEk/ estará sempre atualizado (contenha todas as configurações). Como isso está crescendo (novas opções são adicionadas), não atualizarei todos os violinos anteriores.

Chamadas

  • f("string to image (palette)", "string to image", {object of options});
  • f([[palette pixel], [palette pixel], ..., "string to image", {object of options});

Opções

  • algoritmo: 'balanced' , 'surrounding', 'reverse', 'hsv', 'yiq','lab'
  • velocidade: velocidade de animação
  • movimento: true- a animação deve mostrar movimento da posição inicial ao final
  • surround: se o 'surrounding'algoritmo for selecionado, esse será o peso do ambiente que será levado em consideração no cálculo do peso do pixel fornecido
  • hsv: if 'hsv' algoritmo for selecionado, esses parâmetros controlam quanto matiz, saturação e valor afetam os pesos
  • yiq: se 'qiv' algoritmo for selecionado, esses parâmetros controlam quanto yiq afeta os pesos
  • laboratório: se 'lab' algoritmo for selecionado, esses parâmetros controlam quanto o laboratório afeta os pesos
  • ruído: quanta aleatoriedade será adicionada aos pesos
  • exclusivo: os pixels da paleta devem ser usados ​​apenas uma vez (consulte: Fotomosaicos ou: Quantos programadores são necessários para substituir uma lâmpada? )
  • pixel_1 / pixel_2 {largura, altura}: tamanho do pixel (em pixels: D)

Galeria (para as mostras, estou sempre usando Mona Lisa e American Gothic, a menos que seja especificado outro):


A animação está ótima! mas sua imagem é um pixel menor que o normal.
9788 Calvin's Hobbies

@ Hobbies de Calvin - Tive que cortar em tinta: P Provavelmente é daí que vem a diferença. Atualizada!
eithed

Eu gosto deste: jsfiddle.net/q865W/4
Justin

@Quincunx Cheers! Com a versão ponderada ele funciona ainda melhor
eithed

Uau. 0_0 Isso é muito bom. jsfiddle.net/q865W/6
Justin

24

C, com espaço de cores Lab e pontilhamento aprimorado

Eu disse que tinha terminado? Eu menti. Eu acho que o algoritmo na minha outra solução é o melhor, mas o Perl não é rápido o suficiente para tarefas de processamento de números, então reimplementei meu trabalho no C. Agora ele executa todas as imagens deste post, com uma qualidade mais alta do que o original em cerca de 3 minutos por imagem e a qualidade um pouco mais baixa (nível de 0,5%) é executada em 20 a 30 segundos por imagem. Basicamente, todo o trabalho é feito com o ImageMagick, e o pontilhamento é feito usando a interpolação de spline cúbica do ImageMagick, que fornece um resultado melhor / menos padronizado.

Código

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <math.h>
#include <unistd.h>
#include <wand/MagickWand.h>

#define ThrowWandException(wand) \
{ \
  char \
  *description; \
  \
  ExceptionType \
  severity; \
  \
  description=MagickGetException(wand,&severity); \
  (void) fprintf(stderr,"%s %s %lu %s\n",GetMagickModule(),description); \
  description=(char *) MagickRelinquishMemory(description); \
  abort(); \
  exit(-1); \
}

int width, height; /* Target image size */
MagickWand *source_wand, *target_wand, *img_wand, *target_lab_wand, *img_lab_wand;
PixelPacket *source_pixels, *target_pixels, *img_pixels, *target_lab_pixels, *img_lab_pixels;
Image *img, *img_lab, *target, *target_lab;
CacheView *img_lab_view, *target_lab_view;
ExceptionInfo *e;

MagickWand *load_image(const char *filename) {
  MagickWand *img = NewMagickWand();
  if (!MagickReadImage(img, filename)) {
    ThrowWandException(img);
  }
  return img;
}

PixelPacket *get_pixels(MagickWand *wand) {
  PixelPacket *ret = GetAuthenticPixels(
      GetImageFromMagickWand(wand), 0, 0,
      MagickGetImageWidth(wand), MagickGetImageHeight(wand), e);
  CatchException(e);
  return ret;
}

void sync_pixels(MagickWand *wand) {
  SyncAuthenticPixels(GetImageFromMagickWand(wand), e);
  CatchException(e);
}

MagickWand *transfer_pixels() {
  if (MagickGetImageWidth(source_wand) * MagickGetImageHeight(source_wand)
      != MagickGetImageWidth(target_wand) * MagickGetImageHeight(target_wand)) {
    perror("size mismtch");
  }

  MagickWand *img_wand = CloneMagickWand(target_wand);
  img_pixels = get_pixels(img_wand);
  memcpy(img_pixels, source_pixels, 
      MagickGetImageWidth(img_wand) * MagickGetImageHeight(img_wand) * sizeof(PixelPacket));

  sync_pixels(img_wand);
  return img_wand;
}

MagickWand *image_to_lab(MagickWand *img) {
  MagickWand *lab = CloneMagickWand(img);
  TransformImageColorspace(GetImageFromMagickWand(lab), LabColorspace);
  return lab;
}

int lab_distance(PixelPacket *a, PixelPacket *b) {
  int l_diff = (GetPixelL(a) - GetPixelL(b)) / 256,
      a_diff = (GetPixela(a) - GetPixela(b)) / 256,
      b_diff = (GetPixelb(a) - GetPixelb(b)) / 256;

  return (l_diff * l_diff + a_diff * a_diff + b_diff * b_diff);
}

int should_swap(int x1, int x2, int y1, int y2) {
  int dist = lab_distance(&img_lab_pixels[width * y1 + x1], &target_lab_pixels[width * y1 + x1])
           + lab_distance(&img_lab_pixels[width * y2 + x2], &target_lab_pixels[width * y2 + x2]);
  int swapped_dist = lab_distance(&img_lab_pixels[width * y2 + x2], &target_lab_pixels[width * y1 + x1])
                   + lab_distance(&img_lab_pixels[width * y1 + x1], &target_lab_pixels[width * y2 + x2]);

  return swapped_dist < dist;
}

void pixel_multiply_add(MagickPixelPacket *dest, PixelPacket *src, double mult) {
  dest->red += (double)GetPixelRed(src) * mult;
  dest->green += ((double)GetPixelGreen(src) - 32768) * mult;
  dest->blue += ((double)GetPixelBlue(src) - 32768) * mult;
}

#define min(x,y) (((x) < (y)) ? (x) : (y))
#define max(x,y) (((x) > (y)) ? (x) : (y))

double mpp_distance(MagickPixelPacket *a, MagickPixelPacket *b) {
  double l_diff = QuantumScale * (a->red - b->red),
         a_diff = QuantumScale * (a->green - b->green),
         b_diff = QuantumScale * (a->blue - b->blue);
  return (l_diff * l_diff + a_diff * a_diff + b_diff * b_diff);
}

void do_swap(PixelPacket *pix, int x1, int x2, int y1, int y2) {
  PixelPacket tmp = pix[width * y1 + x1];
  pix[width * y1 + x1] = pix[width * y2 + x2];
  pix[width * y2 + x2] = tmp;
}

int should_swap_dither(double detail, int x1, int x2, int y1, int y2) {
//  const InterpolatePixelMethod method = Average9InterpolatePixel;
  const InterpolatePixelMethod method = SplineInterpolatePixel;

  MagickPixelPacket img1, img2, img1s, img2s, target1, target2;
  GetMagickPixelPacket(img, &img1);
  GetMagickPixelPacket(img, &img2);
  GetMagickPixelPacket(img, &img1s);
  GetMagickPixelPacket(img, &img2s);
  GetMagickPixelPacket(target, &target1);
  GetMagickPixelPacket(target, &target2);
  InterpolateMagickPixelPacket(img, img_lab_view, method, x1, y1, &img1, e);
  InterpolateMagickPixelPacket(img, img_lab_view, method, x2, y2, &img2, e);
  InterpolateMagickPixelPacket(target, target_lab_view, method, x1, y1, &target1, e);
  InterpolateMagickPixelPacket(target, target_lab_view, method, x2, y2, &target2, e);
  do_swap(img_lab_pixels, x1, x2, y1, y2);
//  sync_pixels(img_wand);
  InterpolateMagickPixelPacket(img, img_lab_view, method, x1, y1, &img1s, e);
  InterpolateMagickPixelPacket(img, img_lab_view, method, x2, y2, &img2s, e);
  do_swap(img_lab_pixels, x1, x2, y1, y2);
//  sync_pixels(img_wand);

  pixel_multiply_add(&img1, &img_lab_pixels[width * y1 + x1], detail);
  pixel_multiply_add(&img2, &img_lab_pixels[width * y2 + x2], detail);
  pixel_multiply_add(&img1s, &img_lab_pixels[width * y2 + x2], detail);
  pixel_multiply_add(&img2s, &img_lab_pixels[width * y1 + x1], detail);
  pixel_multiply_add(&target1, &target_lab_pixels[width * y1 + x1], detail);
  pixel_multiply_add(&target2, &target_lab_pixels[width * y2 + x2], detail);

  double dist = mpp_distance(&img1, &target1)
              + mpp_distance(&img2, &target2);
  double swapped_dist = mpp_distance(&img1s, &target1)
                      + mpp_distance(&img2s, &target2);

  return swapped_dist + 1.0e-4 < dist;
}

int main(int argc, char *argv[]) {
  if (argc != 7) {
    fprintf(stderr, "Usage: %s source.png target.png dest nodither_pct dither_pct detail\n", argv[0]);
    return 1;
  }
  char *source_filename = argv[1];
  char *target_filename = argv[2];
  char *dest = argv[3];
  double nodither_pct = atof(argv[4]);
  double dither_pct = atof(argv[5]);
  double detail = atof(argv[6]) - 1;
  const int SWAPS_PER_LOOP = 1000000;
  int nodither_limit = ceil(SWAPS_PER_LOOP * nodither_pct / 100);
  int dither_limit = ceil(SWAPS_PER_LOOP * dither_pct / 100);
  int dither = 0, frame = 0;
  char outfile[256], cmdline[1024];
  sprintf(outfile, "out/%s.png", dest);

  MagickWandGenesis();
  e = AcquireExceptionInfo();
  source_wand = load_image(source_filename);
  source_pixels = get_pixels(source_wand);
  target_wand = load_image(target_filename);
  target_pixels = get_pixels(target_wand);
  img_wand = transfer_pixels();
  img_pixels = get_pixels(img_wand);
  target_lab_wand = image_to_lab(target_wand);
  target_lab_pixels = get_pixels(target_lab_wand);
  img_lab_wand = image_to_lab(img_wand);
  img_lab_pixels = get_pixels(img_lab_wand);
  img = GetImageFromMagickWand(img_lab_wand);
  target = GetImageFromMagickWand(target_lab_wand);
  img_lab_view = AcquireAuthenticCacheView(img, e);
  target_lab_view = AcquireAuthenticCacheView(target,e);
  CatchException(e);

  width = MagickGetImageWidth(img_wand);
  height = MagickGetImageHeight(img_wand);

  while (1) {
    int swaps_made = 0;
    for (int n = 0 ; n < SWAPS_PER_LOOP ; n++) {
      int x1 = rand() % width,
          x2 = rand() % width,
          y1 = rand() % height,
          y2 = rand() % height;

      int swap = dither ?
        should_swap_dither(detail, x1, x2, y1, y2)
        : should_swap(x1, x2, y1, y2);

      if (swap) {
        do_swap(img_pixels, x1, x2, y1, y2);
        do_swap(img_lab_pixels, x1, x2, y1, y2);
        swaps_made ++;
      }
    }

    sync_pixels(img_wand);
    if (!MagickWriteImages(img_wand, outfile, MagickTrue)) {
      ThrowWandException(img_wand);
    }
    img_pixels = get_pixels(img_wand);
    sprintf(cmdline, "cp out/%s.png anim/%s/%05i.png", dest, dest, frame++);
    system(cmdline);

    if (!dither && swaps_made < nodither_limit) {
      sprintf(cmdline, "cp out/%s.png out/%s-nodither.png", dest, dest);
      system(cmdline);
      dither = 1;
    } else if (dither && swaps_made < dither_limit)
      break;
  }

  return 0;
}

Ajuntar com

gcc -std=gnu99 -O3 -march=native -ffast-math \
  -o transfer `pkg-config --cflags MagickWand` \
  transfer.c `pkg-config --libs MagickWand` -lm

Resultados

Principalmente o mesmo que a versão Perl, apenas um pouco melhor, mas existem algumas exceções. O pontilhamento é menos perceptível em geral. Grito -> Noite Estrelada não tem o efeito "montanha flamejante", e o Camaro parece menos glitchy com os pixels cinzentos. Eu acho que o código de espaço de cores da versão Perl tem um bug com pixels de baixa saturação.

Paleta gótica americana

Paleta Mona Lisa

Paleta Noite Estrelada

Paleta de gritos

Paleta de esferas

Mustang (paleta Camaro)

Camaro (paleta Mustang)


Sim senhor, o seu realmente é o melhor por aí. Por que em C gera 0,5% pior?
RMalke

@RMalke Só é pior quando ele só deixa funcionar por 20 a 30 segundos.
trlkly

Poderia, por favor postar os valores que você usou como nodither_pct, dither_pcte detailneste exemplo? Estou executando o seu programa com combinações diferentes, mas, para minhas imagens, elas parecem estar abaixo do ideal e as paletas estão próximas das suas, então ... por favor?
Andreï Kostyrka

@ AndreïKostyrka 0.1 0.1 1.6são os valores que eu usei para gerar essas imagens.
hobbs

@ AndreïKostyrka 0.5 0.5 1.6deve resultar em qualidade quase tão boa com velocidade muito mais rápida.
Hbbs

23

Valor mais próximo do HSL com propagação de erros e pontilhamento

Fiz pequenas adaptações no código que usei para minhas imagens AllRGB . Ele foi projetado para processar imagens de 16 megapixels com restrições de tempo e memória razoáveis, portanto, ele usa algumas classes de estrutura de dados que não estão na biblioteca padrão; no entanto, eu os omiti porque já existe muito código aqui e esse é o código interessante.

Para o AllRGB, eu ajusto manualmente as wavelets que dão prioridade a determinadas áreas da imagem; para esse uso não-guiado, escolho uma wavelet que assume o layout da regra dos terços, colocando o interesse principal em um terço do topo.

Gótico americano com paleta de Mona Lisa Mona Lisa com paleta de American Gothic

O meu favorito dos 36:

Rio com paleta de Mona Lisa

Produto cartesiano completo de (imagem, paleta)

package org.cheddarmonk.graphics;

import org.cheddarmonk.util.*;
import java.awt.Point;
import java.awt.image.*;
import java.io.File;
import java.util.Random;
import javax.imageio.ImageIO;

public class PaletteApproximator {
    public static void main(String[] args) throws Exception {
        // Adjust this to fine-tune for the areas which are most important.
        float[] waveletDefault = new float[] {0.5f, 0.333f, 0.5f, 0.5f, 1};

        generateAndSave(args[0], args[1], args[2], waveletDefault);
    }

    private static void generateAndSave(String paletteFile, String fileIn, String fileOut, float[]... wavelets) throws Exception {
        BufferedImage imgIn = ImageIO.read(new File(fileIn));
        int w = imgIn.getWidth(), h = imgIn.getHeight();

        int[] buf = new int[w * h];
        imgIn.getRGB(0, 0, w, h, buf, 0, w);

        SimpleOctTreeInt palette = loadPalette(paletteFile);
        generate(palette, buf, w, h, wavelets);

        // Masks for R, G, B, A.
        final int[] off = new int[]{0xff0000, 0xff00, 0xff, 0xff000000};
        // The corresponding colour model.
        ColorModel colourModel = ColorModel.getRGBdefault();
        DataBufferInt dbi = new DataBufferInt(buf, buf.length);
        Point origin = new Point(0, 0);
        WritableRaster raster = Raster.createPackedRaster(dbi, w, h, w, off, origin);
        BufferedImage imgOut = new BufferedImage(colourModel, raster, false, null);

        ImageIO.write(imgOut, "PNG", new File(fileOut));
    }

    private static SimpleOctTreeInt loadPalette(String paletteFile) throws Exception {
        BufferedImage img = ImageIO.read(new File(paletteFile));
        int w = img.getWidth(), h = img.getHeight();

        int[] buf = new int[w * h];
        img.getRGB(0, 0, w, h, buf, 0, w);

        // Parameters tuned for 4096x4096
        SimpleOctTreeInt octtree = new SimpleOctTreeInt(0, 1, 0, 1, 0, 1, 16, 12);
        for (int i = 0; i < buf.length; i++) {
            octtree.add(buf[i], transform(buf[i]));
        }

        return octtree;
    }

    private static void generate(SimpleOctTreeInt octtree, int[] buf, int w, int h, float[]... wavelets) {
        int m = w * h;

        LeanBinaryHeapInt indices = new LeanBinaryHeapInt();
        Random rnd = new Random();
        for (int i = 0; i < m; i++) {
            float x = (i % w) / (float)w, y = (i / w) / (float)w;

            float weight = 0;
            for (float[] wavelet : wavelets) {
                weight += wavelet[4] * Math.exp(-Math.pow((x - wavelet[0]) / wavelet[2], 2) - Math.pow((y - wavelet[1]) / wavelet[3], 2));
            }

            // Random element provides some kind of dither
            indices.insert(i, -weight + 0.2f * rnd.nextFloat());
        }

        // Error diffusion buffers.
        float[] errx = new float[m], erry = new float[m], errz = new float[m];

        for (int i = 0; i < m; i++) {
            int idx = indices.pop();
            int x = idx % w, y = idx / w;

            // TODO Bicubic interpolation? For the time being, prefer to scale the input image externally...
            float[] tr = transform(buf[x + w * y]);
            tr[0] += errx[idx]; tr[1] += erry[idx]; tr[2] += errz[idx];

            int pixel = octtree.nearestNeighbour(tr, 2);
            buf[x + y * w] = 0xff000000 | pixel;

            // Don't reuse pixels.
            float[] trPix = transform(pixel);
            boolean ok = octtree.remove(pixel, trPix);
            if (!ok) throw new IllegalStateException("Failed to remove from octtree");

            // Propagate error in 4 directions, not caring whether or not we've already done that pixel.
            // This will lose some error, but that might be a good thing.
            float dx = (tr[0] - trPix[0]) / 4, dy = (tr[1] - trPix[1]) / 4, dz = (tr[2] - trPix[2]) / 4;
            if (x > 0) {
                errx[idx - 1] += dx;
                erry[idx - 1] += dy;
                errz[idx - 1] += dz;
            }
            if (x < w - 1) {
                errx[idx + 1] += dx;
                erry[idx + 1] += dy;
                errz[idx + 1] += dz;
            }
            if (y > 0) {
                errx[idx - w] += dx;
                erry[idx - w] += dy;
                errz[idx - w] += dz;
            }
            if (y < h - 1) {
                errx[idx + w] += dx;
                erry[idx + w] += dy;
                errz[idx + w] += dz;
            }
        }
    }

    private static final float COS30 = (float)Math.sqrt(3) / 2;
    private static float[] transform(int rgb) {
        float r = ((rgb >> 16) & 0xff) / 255.f;
        float g = ((rgb >> 8) & 0xff) / 255.f;
        float b = (rgb & 0xff) / 255.f;

        // HSL cone coords
        float cmax = (r > g) ? r : g; if (b > cmax) cmax = b;
        float cmin = (r < g) ? r : g; if (b < cmin) cmin = b;
        float[] cone = new float[3];
        cone[0] = (cmax + cmin) / 2;
        cone[1] = 0.5f * (1 + r - (g + b) / 2);
        cone[2] = 0.5f * (1 + (g - b) * COS30);
        return cone;
    }
}

22

Pitão

Não é bonito código, nem por resultados.

from blist import blist
from PIL import Image
import random

def randpop(colors):
    j = random.randrange(len(colors))
    return colors.pop(j)

colors = blist(Image.open('in1.png').getdata())
random.shuffle(colors)
target = Image.open('in2.png')

out = target.copy()
data = list(list(i) for i in out.getdata())

assert len(data) == len(colors)

w, h = out.size

coords = []
for i in xrange(h):
    for j in xrange(w):
        coords.append((i, j))

# Adjust color balance
dsum = [sum(d[i] for d in data) for i in xrange(3)]
csum = [sum(c[i] for c in colors) for i in xrange(3)]
adjust = [(csum[i] - dsum[i]) // len(data) for i in xrange(3)]
for i, j in coords:
    for k in xrange(3):
        data[i*w + j][k] += adjust[k]

random.shuffle(coords)

# larger value here gives better results but take longer
choose = 100
threshold = 10

done = set()
while len(coords):
    if not len(coords) % 1000:
        print len(coords) // 1000
    i, j = coords.pop()
    ind = i*w + j
    done.add(ind)
    t = data[ind]
    dmin = 255*3
    kmin = 0
    choices = []
    while colors and len(choices) < choose:
        k = len(choices)
        choices.append(randpop(colors))
        c = choices[-1]
        d = sum(abs(t[l] - c[l]) for l in xrange(3))
        if d < dmin:
            dmin = d
            kmin = k
            if d < threshold:
                break
    c = choices.pop(kmin)
    data[ind] = c
    colors.extend(choices)

    # Push the error to nearby pixels for dithering
    if ind + 1 < len(data) and ind + 1 not in done:
        ind2 = ind + 1
    elif ind + w < len(data) and ind + w not in done:
        ind2 = ind + w
    elif ind > 0 and ind - 1 not in done:
        ind2 = ind - 1
    elif ind - w > 0 and ind - w not in done:
        ind2 = ind - w
    else:
        ind2 = None
    if ind2 is not None:
        for k in xrange(3):
            err = abs(t[k] - c[k])
            data[ind2][k] += err

out.putdata(data)
out.save('out.png')

Possíveis melhorias:

  • correção de cores mais inteligente?
  • métrica de melhor qualidade?
  • envie o erro para todos os pixels circundantes em vez de um

Feio (1-> 2): 1-> 2

Um pouco melhor (2-> 1): 2-> 1

Decente (2-> 3): 2-> 3

Como um traçador de raios ruim (3-> 4): 3-> 4

Trapaça - use todos os pixels bons na metade superior e diga que a tinta acabou: 1-> 2


3
O último é ... uma ideia interessante. Mas ainda não votando.
John Dvorak

20

Python (usando kd-tree e luminosidade)

Bom desafio. Eu decidi seguir uma abordagem kd-tree. Portanto, a idéia básica por trás de uma abordagem kd-tree é que ela divide as cores e a luminosidade de acordo com a presença na figura.

Portanto, para a árvore kd, o primeiro tipo é baseado no vermelho. Ele divide todas as cores em dois grupos aproximadamente iguais de vermelho (vermelho claro e vermelho escuro). Em seguida, divide essas duas partições ao longo dos verdes. Em seguida azul e depois luminosidade e depois vermelho novamente. E assim por diante até que a árvore seja construída. Nesta abordagem, construí uma árvore kd para a imagem de origem e a imagem de destino. Depois disso, mapeei a árvore da origem para o destino e substitui os dados de cores do arquivo de destino. Todos os resultados são mostrados aqui .

Alguns exemplos:

Mona Lisa -> Gótico Americano

Monalisa gótico americano (estilo mona_lisa)

Gótico Americano -> Mona Lisa

gótico americano mona_lisa (estilo gótico americano)

Noite Estrelada -> O Grito

noite estrelada grito estrelado

O Grito -> Noite Estrelada

grito estrelas gritantes

Esferas do arco-íris

insira a descrição da imagem aqui mona lisa balls bolas gritantes

Aqui está um curta-metragem usando o criador de molduras para filmes @ Calvin's Hobbies:

insira a descrição da imagem aqui

E agora para o código :-)

from PIL import Image

""" Computation of hue, saturation, luminosity.
Based on http://stackoverflow.com/questions/3732046/how-do-you-get-the-hue-of-a-xxxxxx-colour
"""
def rgbToLsh(t):
    r = t[0]
    g = t[1]
    b = t[2]
    r /= 255.
    g /= 255.
    b /= 255.
    vmax = max([r, g, b])
    vmin = min([r, g, b]);
    h = s = l = (vmax + vmin) / 2.;

    if (vmax == vmin):
        h = s = 0.  # achromatic
    else:
        d = vmax - vmin;
        if l > 0.5:
            s = d / (2. - vmax - vmin)
        else:
            s = d / (vmax + vmin);
        if vmax == r:
            if g<b: 
                m = 6. 
            else: 
                m = 0. 
            h = (g - b) / d + m
        elif vmax == g: 
            h = (b - r) / d + 2.
        elif vmax == b: 
            h = (r - g) / d + 4.
        h /= 6.;
    return [l,s,h];



""" KDTree implementation.
Based on https://code.google.com/p/python-kdtree/ 
"""
__version__ = "1r11.1.2010"
__all__ = ["KDTree"]

def square_distance(pointA, pointB):
    # squared euclidean distance
    distance = 0
    dimensions = len(pointA) # assumes both points have the same dimensions
    for dimension in range(dimensions):
        distance += (pointA[dimension] - pointB[dimension])**2
    return distance

class KDTreeNode():
    def __init__(self, point, left, right):
        self.point = point
        self.left = left
        self.right = right

    def is_leaf(self):
        return (self.left == None and self.right == None)

class KDTreeNeighbours():
    """ Internal structure used in nearest-neighbours search.
    """
    def __init__(self, query_point, t):
        self.query_point = query_point
        self.t = t # neighbours wanted
        self.largest_distance = 0 # squared
        self.current_best = []

    def calculate_largest(self):
        if self.t >= len(self.current_best):
            self.largest_distance = self.current_best[-1][1]
        else:
            self.largest_distance = self.current_best[self.t-1][1]

    def add(self, point):
        sd = square_distance(point, self.query_point)
        # run through current_best, try to find appropriate place
        for i, e in enumerate(self.current_best):
            if i == self.t:
                return # enough neighbours, this one is farther, let's forget it
            if e[1] > sd:
                self.current_best.insert(i, [point, sd])
                self.calculate_largest()
                return
        # append it to the end otherwise
        self.current_best.append([point, sd])
        self.calculate_largest()

    def get_best(self):
        return [element[0] for element in self.current_best[:self.t]]



class KDTree():
    """ KDTree implementation.

        Example usage:

            from kdtree import KDTree

            data = <load data> # iterable of points (which are also iterable, same length)
            point = <the point of which neighbours we're looking for>

            tree = KDTree.construct_from_data(data)
            nearest = tree.query(point, t=4) # find nearest 4 points
    """

    def __init__(self, data):

        self.data_listing = []
        def build_kdtree(point_list, depth):

            # code based on wikipedia article: http://en.wikipedia.org/wiki/Kd-tree
            if not point_list:
                return None

            # select axis based on depth so that axis cycles through all valid values
            axis = depth % 4 #len(point_list[0]) # assumes all points have the same dimension

            # sort point list and choose median as pivot point,
            # TODO: better selection method, linear-time selection, distribution
            point_list.sort(key=lambda point: point[axis])
            median = len(point_list)/2 # choose median

            # create node and recursively construct subtrees
            node = KDTreeNode(point=point_list[median],
                              left=build_kdtree(point_list[0:median], depth+1),
                              right=build_kdtree(point_list[median+1:], depth+1))

            # add point to listing                   
            self.data_listing.append(point_list[median])
            return node

        self.root_node = build_kdtree(data, depth=0)

    @staticmethod
    def construct_from_data(data):
        tree = KDTree(data)
        return tree

    def query(self, query_point, t=1):
        statistics = {'nodes_visited': 0, 'far_search': 0, 'leafs_reached': 0}

        def nn_search(node, query_point, t, depth, best_neighbours):
            if node == None:
                return

            #statistics['nodes_visited'] += 1

            # if we have reached a leaf, let's add to current best neighbours,
            # (if it's better than the worst one or if there is not enough neighbours)
            if node.is_leaf():
                #statistics['leafs_reached'] += 1
                best_neighbours.add(node.point)
                return

            # this node is no leaf

            # select dimension for comparison (based on current depth)
            axis = depth % len(query_point)

            # figure out which subtree to search
            near_subtree = None # near subtree
            far_subtree = None # far subtree (perhaps we'll have to traverse it as well)

            # compare query_point and point of current node in selected dimension
            # and figure out which subtree is farther than the other
            if query_point[axis] < node.point[axis]:
                near_subtree = node.left
                far_subtree = node.right
            else:
                near_subtree = node.right
                far_subtree = node.left

            # recursively search through the tree until a leaf is found
            nn_search(near_subtree, query_point, t, depth+1, best_neighbours)

            # while unwinding the recursion, check if the current node
            # is closer to query point than the current best,
            # also, until t points have been found, search radius is infinity
            best_neighbours.add(node.point)

            # check whether there could be any points on the other side of the
            # splitting plane that are closer to the query point than the current best
            if (node.point[axis] - query_point[axis])**2 < best_neighbours.largest_distance:
                #statistics['far_search'] += 1
                nn_search(far_subtree, query_point, t, depth+1, best_neighbours)

            return

        # if there's no tree, there's no neighbors
        if self.root_node != None:
            neighbours = KDTreeNeighbours(query_point, t)
            nn_search(self.root_node, query_point, t, depth=0, best_neighbours=neighbours)
            result = neighbours.get_best()
        else:
            result = []

        #print statistics
        return result


#List of files: 
files = ['JXgho.png','N6IGO.png','c5jq1.png','itzIe.png','xPAwA.png','y2VZJ.png']

#Loop over source files 
for im_orig in range(len(files)):
    srch = Image.open(files[im_orig])   #Open file handle 
    src = srch.load();                  #Load file  

    # Build data structure (R,G,B,lum,xpos,ypos) for source file
    srcdata =  [(src[i,j][0],src[i,j][1],src[i,j][2],rgbToLsh(src[i,j])[0],i,j) \
                     for i in range(srch.size[0]) \
                     for j in range(srch.size[1])]  

    # Build kd-tree for source
    srctree = KDTree.construct_from_data(srcdata)

    for im in range(len(files)):
        desh = Image.open(files[im])
        des = desh.load();

        # Build data structure (R,G,B,lum,xpos,ypos) for destination file
        desdata =  [(des[i,j][0],des[i,j][1],des[i,j][2],rgbToLsh(des[i,j]),i,j) \
                     for i in range(desh.size[0]) \
                     for j in range(desh.size[1])]  

        # Build kd-tree for destination
        destree = KDTree.construct_from_data(desdata)

        # Switch file mode
        desh.mode = srch.mode
        for k in range(len(srcdata)):
            # Get locations from kd-tree sorted data
            i   = destree.data_listing[k][-2]
            j   = destree.data_listing[k][-1]
            i_s = srctree.data_listing[k][-2]
            j_s = srctree.data_listing[k][-1]

            # Overwrite original colors with colors from source file 
            des[i,j] = src[i_s,j_s]

        # Save to disk  
        desh.save(files[im_orig].replace('.','_'+`im`+'.'))

Eu não percebi isso há um ano, mas é muito bom!
Hbbs

16

Pitão

Apenas para manter a bola rolando, aqui está minha própria resposta simples e dolorosamente lenta.

import Image

def countColors(image):
    colorCounts = {}
    for color in image.getdata():
        if color in colorCounts:
            colorCounts[color] += 1
        else:
            colorCounts[color] = 1
    return colorCounts

def colorDist(c1, c2):
    def ds(c1, c2, i):
        return (c1[i] - c2[i])**2
    return (ds(c1, c2, 0) + ds(c1, c2, 1) + ds(c1, c2, 2))**0.5

def findClosestColor(palette, color):
    closest = None
    minDist = (3*255**2)**0.5
    for c in palette:
        dist = colorDist(color, c)
        if dist < minDist:
            minDist = dist
            closest = c
    return closest

def removeColor(palette, color):
    if palette[color] == 1:
        del palette[color]
    else:
        palette[color] -= 1

def go(paletteFile, sourceFile):
    palette = countColors(Image.open(paletteFile).convert('RGB'))
    source = Image.open(sourceFile).convert('RGB')
    copy = Image.new('RGB', source.size)
    w, h = copy.size

    for x in range(w):
        for y in range(h):
            c = findClosestColor(palette, source.getpixel((x, y)))
            removeColor(palette, c)
            copy.putpixel((x, y), c)
        print x #print progress
    copy.save('copy.png')

#the respective file paths go here
go('../ag.png', '../r.png')

Para cada pixel na origem, ele procura o pixel não utilizado na paleta mais próximo do cubo de cores RGB. É basicamente o mesmo que o algoritmo de Quincunx, mas sem aleatoriedade e uma função de comparação de cores diferente.

Você pode dizer que eu passo da esquerda para a direita, pois o lado direito da imagem tem muito menos detalhes devido ao esgotamento de cores semelhantes.

Rio do gótico americano

Rio do gótico americano

Mona Lisa from Esferas do Arco-Íris

Mona Lisa from Esferas do Arco-Íris


1
Sra. Lisa é um pouco amarelada ...
tomsmeding

4
Eu realmente gosto de transição no Rio de American Gothic de 'agradável' esquerda para a direita 'abstrata' =)
flawr

12

Haskell

Eu tentei algumas abordagens diferentes usando pesquisas de vizinhos mais próximos antes de decidir sobre esta solução (que foi realmente minha primeira ideia). Primeiro converto os formatos de pixel das imagens em YCbCr e construo duas listas contendo seus dados de pixel. Em seguida, as listas são classificadas, dando precedência ao valor da luminância. Depois disso, basta substituir a lista de pixels classificados da imagem de entrada pela imagem da paleta e, em seguida, recorrer à ordem original e usá-la para construir uma nova imagem.

module Main where

import System.Environment    (getArgs)
import System.Exit           (exitSuccess, exitFailure)
import System.Console.GetOpt (getOpt, ArgOrder(..), OptDescr(..), ArgDescr(..))
import Data.List             (sortBy)

import Codec.Picture
import Codec.Picture.Types

import qualified Data.Vector as V

main :: IO ()
main = do
    (ioOpts, _) <- getArgs >>= getOpts
    opts        <- ioOpts
    image       <- loadImage $ imageFile opts
    palette     <- loadImage $ paletteFile opts
    case swapPalette image palette of
      Nothing -> do
          putStrLn "Error: image and palette dimensions do not match"
          exitFailure
      Just img ->
          writePng (outputFile opts) img

swapPalette :: Image PixelYCbCr8 -> Image PixelYCbCr8 -> Maybe (Image PixelRGB8)
swapPalette img pal
    | area1 == area2 =
        let cmpCr (_, (PixelYCbCr8 _ _ r1)) (_, (PixelYCbCr8 _ _ r2)) = r1 `compare` r2
            cmpCb (_, (PixelYCbCr8 _ c1 _)) (_, (PixelYCbCr8 _ c2 _)) = c1 `compare` c2
            cmpY  (_, (PixelYCbCr8 y1 _ _)) (_, (PixelYCbCr8 y2 _ _)) = y2 `compare` y1
            w       = imageWidth  img
            h       = imageHeight img
            imgData = sortBy cmpY $ sortBy cmpCr $ sortBy cmpCb $ zip [1 :: Int ..] $ getPixelList img
            palData = sortBy cmpY $ sortBy cmpCr $ sortBy cmpCb $ zip [1 :: Int ..] $ getPixelList pal
            newData = zipWith (\(n, _) (_, p) -> (n, p)) imgData palData
            pixData = map snd $ sortBy (\(n1, _) (n2, _) -> n1 `compare` n2) newData
            dataVec = V.reverse $ V.fromList pixData
        in  Just $ convertImage $ generateImage (lookupPixel dataVec w h) w h
    | otherwise = Nothing
    where area1 = (imageWidth img) * (imageHeight img)
          area2 = (imageWidth pal) * (imageHeight pal)

lookupPixel :: V.Vector PixelYCbCr8 -> Int -> Int -> Int -> Int -> PixelYCbCr8
lookupPixel vec w h x y = vec V.! i
    where i = flattenIndex w h x y

getPixelList :: Image PixelYCbCr8 -> [PixelYCbCr8]
getPixelList img = foldl (\ps (x, y) -> (pixelAt img x y):ps) [] coords
    where coords = [(x, y) | x <- [0..(imageWidth img) - 1], y <- [0..(imageHeight img) - 1]]

flattenIndex :: Int -> Int -> Int -> Int -> Int
flattenIndex _ h x y = y + (x * h)

-------------------------------------------------
-- Command Line Option Functions
-------------------------------------------------

getOpts :: [String] -> IO (IO Options, [String])
getOpts args = case getOpt Permute options args of
    (opts, nonOpts, []) -> return (foldl (>>=) (return defaultOptions) opts, nonOpts)
    (_, _, errs)        -> do
        putStrLn $ concat errs
        printUsage
        exitFailure

data Options = Options
  { imageFile   :: Maybe FilePath
  , paletteFile :: Maybe FilePath
  , outputFile  :: FilePath
  }

defaultOptions :: Options
defaultOptions = Options
  { imageFile   = Nothing
  , paletteFile = Nothing
  , outputFile  = "out.png"
  }

options :: [OptDescr (Options -> IO Options)]
options = [ Option ['i'] ["image"]   (ReqArg setImage   "FILE") "",
            Option ['p'] ["palette"] (ReqArg setPalette "FILE") "",
            Option ['o'] ["output"]  (ReqArg setOutput  "FILE") "",
            Option ['v'] ["version"] (NoArg showVersion)        "",
            Option ['h'] ["help"]    (NoArg exitPrintUsage)     ""]

setImage :: String -> Options -> IO Options
setImage image opts = return $ opts { imageFile = Just image }

setPalette :: String -> Options -> IO Options
setPalette palette opts = return $ opts { paletteFile = Just palette }

setOutput :: String -> Options -> IO Options
setOutput output opts = return $ opts { outputFile = output }

printUsage :: IO ()
printUsage = do
    putStrLn "Usage: repix [OPTION...] -i IMAGE -p PALETTE [-o OUTPUT]"
    putStrLn "Rearrange pixels in the palette file to closely resemble the given image."
    putStrLn ""
    putStrLn "-i, --image    specify the image to transform"
    putStrLn "-p, --palette  specify the image to use as the palette"
    putStrLn "-o, --output   specify the output image file"
    putStrLn ""
    putStrLn "-v, --version  display version information and exit"
    putStrLn "-h, --help     display this help and exit"

exitPrintUsage :: a -> IO Options
exitPrintUsage _ = do
    printUsage
    exitSuccess

showVersion :: a -> IO Options
showVersion _ = do
    putStrLn "Pixel Rearranger v0.1"
    exitSuccess

-------------------------------------------------
-- Image Loading Util Functions
-------------------------------------------------

loadImage :: Maybe FilePath -> IO (Image PixelYCbCr8)
loadImage Nothing     = do
    printUsage
    exitFailure
loadImage (Just path) = do
    rdImg <- readImage path
    case rdImg of
      Left err -> do
          putStrLn err
          exitFailure
      Right img -> getRGBImage img

getRGBImage :: DynamicImage -> IO (Image PixelYCbCr8)
getRGBImage dynImg =
    case dynImg of
      ImageYCbCr8 img -> return img
      ImageRGB8   img -> return $ convertImage img
      ImageY8     img -> return $ convertImage (promoteImage img :: Image PixelRGB8)
      ImageYA8    img -> return $ convertImage (promoteImage img :: Image PixelRGB8)
      ImageCMYK8  img -> return $ convertImage (convertImage img :: Image PixelRGB8)
      ImageRGBA8  img -> return $ convertImage (pixelMap dropTransparency img :: Image PixelRGB8)
      _               -> do
          putStrLn "Error: incompatible image type."
          exitFailure

Resultados

As imagens que meu programa produz tendem a ser menos vivas do que muitas outras soluções, e não lidam bem com grandes áreas sólidas ou gradientes.

Aqui está um link para o álbum completo.

Gótico Americano -> Mona Lisa

Mona Lisa -> Gótico Americano

Esferas -> Mona Lisa

O Grito -> Noite Estrelada

O Grito -> Esferas


3
Eu gosto do pontilhamento em (Spheres -> Mona Lisa), mas de onde são esses artefatos feios (Scream -> Spheres)?
John Dvorak

1
Os artefatos são um efeito colateral de como meu algoritmo classifica os pixels. No momento, a diferença de vermelho de cada pixel tem precedência sobre a diferença de azul na etapa de classificação, o que significa que cores semelhantes na imagem de entrada podem ser combinadas com cores muito diferentes da imagem da paleta. No entanto, estou quase certo de que esse mesmo efeito é o que causa o aparente tremor em imagens como as Esferas -> Mona Lisa, então decidi mantê-lo.
ChaseC

9

Java

Inspirado na resposta java anterior do Quincunx

     package paletteswap;

import java.awt.Point;
import java.awt.image.BufferedImage;
import java.io.File;
import java.io.IOException;
import java.util.ArrayList;
import java.util.BitSet;
import java.util.Collections;
import java.util.Comparator;
import java.util.List;

import javax.imageio.ImageIO;

public class Test
{
    public static class Bits
    {

        public static BitSet convert( int value )
        {
            BitSet bits = new BitSet();
            int index = 0;
            while ( value != 0L )
            {
                if ( value % 2 != 0 )
                {
                    bits.set( index );
                }
                ++index;
                value = value >>> 1;
            }
            return bits;
        }

        public static int convert( BitSet bits )
        {
            int value = 0;
            for ( int i = 0; i < bits.length(); ++i )
            {
                value += bits.get( i ) ? ( 1 << i ) : 0;
            }
            return value;
        }
    }

    public static void main( String[] args ) throws IOException
    {
        BufferedImage source = ImageIO.read( resource( "river.png" ) ); // My names
                                                                            // for the
                                                                            // files
        BufferedImage palette = ImageIO.read( resource( "farmer.png" ) );
        BufferedImage result = rearrange( source, palette );
        ImageIO.write( result, "png", resource( "result.png" ) );
    }

    public static BufferedImage rearrange( BufferedImage source, BufferedImage palette )
    {
        BufferedImage result = new BufferedImage( source.getWidth(), source.getHeight(), BufferedImage.TYPE_INT_RGB );

        // This creates a list of points in the Source image.
        // Then, we shuffle it and will draw points in that order.
        List<Point> samples = getPoints( source.getWidth(), source.getHeight() );
        Collections.sort( samples, new Comparator<Point>()
        {

            @Override
            public int compare( Point o1, Point o2 )
            {
                int c1 = getRGB( source, o1.x, o1.y );
                int c2 = getRGB( source, o2.x, o2.y );
                return c1 -c2;
            }
        } );

        // Create a list of colors in the palette.
        List<Integer> colors = getColors( palette );

        while ( !samples.isEmpty() )
        {
            Point currentPoint = samples.remove( 0 );
            int sourceAtPoint = getRGB( source, currentPoint.x, currentPoint.y );
            int colorIndex = binarySearch( colors, sourceAtPoint );
            int bestColor = colors.remove( colorIndex );
            setRGB( result, currentPoint.x, currentPoint.y, bestColor );
        }
        return result;
    }

    public static int unpack( int rgbPacked )
    {
        BitSet packed = Bits.convert( rgbPacked );
        BitSet rgb = Bits.convert( 0 );
        for (int i=0; i<8; i++)
        {
            rgb.set( i,    packed.get( i*3 )  );
            rgb.set( i+16,    packed.get( i*3+1 )  );
            rgb.set( i+8,    packed.get( i*3+2 )  );
        }
        return Bits.convert( rgb);
    }

    public static int pack( int rgb )
    {
        int myrgb = rgb & 0x00FFFFFF;

        BitSet bits = Bits.convert( myrgb );
        BitSet packed = Bits.convert( 0 );

        for (int i=0; i<8; i++)
        {
            packed.set( i*3,    bits.get( i )  );
            packed.set( i*3+1,  bits.get( i+16 )  );
            packed.set( i*3+2,  bits.get( i+8 )  );
        }
        return Bits.convert( packed);

    }

    public static int getRGB( BufferedImage image, int x, int y )
    {
        return pack( image.getRGB( x, y ) );
    }

    public static void setRGB( BufferedImage image, int x, int y, int color )
    {
        image.setRGB( x, y, unpack( color ) );
    }

    public static List<Point> getPoints( int width, int height )
    {
        List<Point> points = new ArrayList<>( width * height );
        for ( int x = 0; x < width; x++ )
        {
            for ( int y = 0; y < height; y++ )
            {
                points.add( new Point( x, y ) );
            }
        }
        return points;
    }

    public static List<Integer> getColors( BufferedImage img )
    {
        int width = img.getWidth();
        int height = img.getHeight();
        List<Integer> colors = new ArrayList<>( width * height );
        for ( int x = 0; x < width; x++ )
        {
            for ( int y = 0; y < height; y++ )
            {
                colors.add( getRGB( img, x, y ) );
            }
        }
        Collections.sort( colors );
        return colors;
    }

    public static int binarySearch( List<Integer> toSearch, int obj )
    {
        int index = toSearch.size() >> 1;
        for ( int guessChange = toSearch.size() >> 2; guessChange > 0; guessChange >>= 1 )
        {
            int value = toSearch.get( index );
            if ( obj == value )
            {
                return index;
            }
            else if ( obj < value )
            {
                index -= guessChange;
            }
            else
            {
                index += guessChange;
            }
        }
        return index;
    }

    public static File resource( String fileName )
    { // This method is here solely
        // for my ease of use (I put
        // the files under <Project
        // Name>/Resources/ )
        return new File( System.getProperty( "user.home" ) + "/pictureswap/" + fileName );
    }
}

Mona lisa -> Agricultores

insira a descrição da imagem aqui

O que é que classifica os pontos que precisam ser substituídos por sua intensidade, em vez de aleatoriamente.


8

Rubi

Visão global:

Abordagem realmente simples, mas parece obter bons resultados:

  1. Pegue a paleta e o alvo, classifique os pixels por alguma função; chame essas matrizes de "referência". Eu escolhi classificar por HSLA, mas preferindo Luminância a Saturação a Matiz (também conhecido como "LSHA")
  2. Crie a imagem de saída iterando sobre cada pixel da imagem de destino, localizando para onde ela é classificada na matriz de referência de destino e pegando o pixel da paleta que foi classificada para o mesmo índice na matriz de referência da paleta.

Código:

require 'rubygems'
require 'chunky_png'
require 'rmagick' # just for the rgba => hsla converter, feel free to use something lighter-weight you have on hand

def pixel_array_for_image(image)
  # [r, b, g, a]
  image.pixels.map{|p| ChunkyPNG::Color.to_truecolor_alpha_bytes(p)}
end

def sorted_pixel_references(pixel_array)
  pixel_array.map{|a| yield(a)}.map.with_index.sort_by(&:first).map(&:last)
end

def sort_by_lsha(pixel_array)
  sorted_pixel_references(pixel_array) {|p|
    # feel free to drop in any sorting function you want here!
    hsla = Magick::Pixel.new(*p).to_hsla # [h, s, l, a]
    [hsla[2], hsla[1], hsla[0], hsla[3]]
  }
end

def make_target_out_of_palette(target_filename, palette_filename, output_filename)
  puts "making #{target_filename} out of #{palette_filename}"

  palette = ChunkyPNG::Image.from_file(palette_filename)
  target = ChunkyPNG::Image.from_file(target_filename)
  puts "  loaded images"

  palette_array = pixel_array_for_image(palette)
  target_array = pixel_array_for_image(target)
  puts "  have pixel arrays"

  palette_spr = sort_by_lsha(palette_array)
  target_spr = sort_by_lsha(target_array)
  puts "  have sorted-pixel reference arrays"

  output = ChunkyPNG::Image.new(target.dimension.width, target.dimension.height, ChunkyPNG::Color::TRANSPARENT)
  (0...target_array.count).each { |index|
    spr_index = target_spr.index(index)
    index_in_palette = palette_spr[spr_index]
    palette_pixel = palette_array[index_in_palette]
    index_as_x = (index % target.dimension.width)
    index_as_y = (index / target.dimension.width)
    output[index_as_x, index_as_y] = ChunkyPNG::Color.rgba(*palette_pixel)
  }
  output.save(output_filename)
  puts "  saved to #{output_filename}"
end

palette_filename, target_filename, output_filename = ARGV
make_target_out_of_palette(target_filename, palette_filename, output_filename)

Resultados:

http://imgur.com/a/Iu7Ds

Luzes:

Noite estrelada feita de grito Gótico americano feito de Mona Lisa Mona Lisa feita a partir do rio A foto do rio feita de Noite Estrelada


2
Você pode adicionar as paletas de origem para cada imagem?
PlasmaHH

7

Perl

Aqui está uma abordagem bastante simplista. Demora cerca de cinco segundos para gerar 100 quadros por par de imagens no meu MacBook Pro com um espaço de memória de cerca de 120 MB.

A idéia é classificar os pixels nas imagens e na paleta por RGB compactado em 24 bits e substituir as cores na fonte pelas cores da paleta sequencialmente.

#!/usr/bin/env perl

use 5.020; # just because
use strict;
use warnings;

use Const::Fast;
use GD;
GD::Image->trueColor(1);

use Path::Class;

const my $COLOR => 0;
const my $COORDINATES => 1;
const my $RGB => 2;
const my $ANIMATION_FRAMES => 100;

const my %MASK => (
    RED => 0x00ff0000,
    GREEN => 0x0000ff00,
    BLUE => 0x000000ff,
);

run(@ARGV);

sub run {
    unless (@_ == 2) {
        die "Need source and palette images\n";
    }
    my $source_file = file(shift)->resolve;
    my $palette_file = file(shift)->resolve;

    my $source = GD::Image->new("$source_file")
        or die "Failed to create source image from '$source_file'";
    my $palette = GD::Image->new("$palette_file")
        or die "Failed to create palette image from '$palette_file'";

    my %source =  map { $_ => $source->$_ } qw(width height);
    my %palette = map { $_ => $palette->$_ } qw(width height);
    my ($frame_prefix) = ($source_file->basename =~ /\A([^.]+)/);

    unless (
        (my $source_area = $source{width} * $source{height}) <=
        (my $palette_area = $palette{width} * $source{height})
    ) {
        die "Source area ($source_area) is greater than palette area ($palette_area)";
    }

    my ($last_frame, $png) = recreate_source_image_from_palette(
        \%source,
        get_source_pixels( get_pixels_by_color($source, \%source) ),
        get_palette_colors( get_pixels_by_color($palette, \%palette) ),
        sub { save_frame($frame_prefix, @_) }
    );

    save_frame($frame_prefix, $last_frame, $png);
    return;
}

sub save_frame {
    my $frame_prefix = shift;
    my $frame = shift;
    my $png = shift;
    file(
        sprintf("${frame_prefix}-%d.png", $frame)
    )->spew(iomode => '>:raw', $$png);
    return;
}

sub recreate_source_image_from_palette {
    my $dim = shift;
    my $source_pixels = shift;
    my $palette_colors = shift;
    my $callback = shift;
    my $frame = 0;

    my %colors;
    $colors{$_} = undef for @$palette_colors;

    my $gd = GD::Image->new($dim->{width}, $dim->{height}, 1);
    for my $x (keys %colors) {
          $colors{$x} = $gd->colorAllocate(unpack_rgb($x));
    }

    my $period = sprintf '%.0f', @$source_pixels / $ANIMATION_FRAMES;
    for my $i (0 .. $#$source_pixels) {
        $gd->setPixel(
            @{ $source_pixels->[$i] },
            $colors{ $palette_colors->[$i] }
        );
        if ($i % $period == 0) {
            $callback->($frame, \ $gd->png);
            $frame += 1;
        }
    }
    return ($frame, \ $gd->png);
}

sub get_palette_colors { [ map sprintf('%08X', $_->[$COLOR]), @{ $_[0] } ] }

sub get_source_pixels { [ map $_->[$COORDINATES], @{ $_[0] } ] }

sub get_pixels_by_color {
    my $gd = shift;
    my $dim = shift;
    return [
        sort { $a->[$COLOR] <=> $b->[$COLOR] }
        map {
            my $y = $_;
            map {
                [ pack_rgb( $gd->rgb( $gd->getPixel($_, $y) ) ), [$_, $y] ];
            } 0 .. $dim->{width}
        } 0 .. $dim->{height}
    ];
}

sub pack_rgb { $_[0] << 16 | $_[1] << 8 | $_[2] }

sub unpack_rgb {
    my ($r, $g, $b) = map $MASK{$_} & hex($_[0]), qw(RED GREEN BLUE);
    return ($r >> 16, $g >> 8, $b);
}

Resultado

Gritar usando a paleta Noite Estrelada

Gritar usando a paleta Noite Estrelada

Gótico americano usando cores da Mona Lisa

Gótico americano usando cores da Mona Lisa

Mona Lisa usando cores Scream

Mona Lisa usando cores Scream

Rio usando cores de mármores

Rio usando cores de mármores

Animações

Como eu era preguiçosa, coloquei as animações no YouTube: Mona Lisa usando cores de Starry Night e American Gothic usando cores de Mona Lisa .


7

Pitão

Achei que eu aproveitaria essa pequena oportunidade para pegar o código de golfe e usá-lo como uma desculpa para trabalhar em minhas costeletas do Python, já que ele vem aparecendo com mais frequência no trabalho atualmente. Corri através de alguns algoritmos, incluindo alguns com tempo O (n ^ 2) e O (nlog (n)) para tentar otimizar as cores, mas ficou muito claro que isso era computacionalmente caro e na verdade tinha muito pouco efeito sobre o resultado aparente. Então, a seguir, é feita uma análise sobre as coisas que funcionam no tempo O (n) (basicamente instantaneamente no meu sistema), que obtém o elemento visual mais importante (luminância) o mais correto possível e permite que o croma chegue aonde for.

from PIL import Image
def check(palette, copy):
    palette = sorted(palette.getdata())
    copy = sorted(copy.getdata())
    print "Master says it's good!" if copy == palette else "The master disapproves."

def GetLuminance(pixel):
    # Extract the pixel channel data
    b, g, r = pixel
    # and used the standard luminance curve to get luminance.
    return .3*r+.59*g+.11*b

print "Putting pixels on the palette..."
# Open up the image and get all of the pixels out of it. (Memory intensive!)
palette = Image.open("2.png").convert(mode="RGB")

pixelsP = [] # Allocate the array
width,height = palette.size # Unpack the image size
for y in range(height): # Scan the lines
    for x in range(width): # within the line, scan the pixels
        curpixel = palette.getpixel((x,y)) # get the pixel
        pixelsP.append((GetLuminance(curpixel),curpixel)) # and add a (luminance, color) tuple to the array.


# sort the pixels by the calculated luminescence
pixelsP.sort()

print "Getting the reference picture..."
# Open up the image and get all of the pixels out of it. (Memory intensive!)
source = Image.open("6.png").convert(mode="RGB")
pixelsR = [] # Allocate the array
width,height = source.size # Unpack the image size
for y in range(height): # Scan the lines
    for x in range(width): # within the line, scan the pixels
        curpixel = source.getpixel((x,y)) # get the pixel
        pixelsR.append((GetLuminance(curpixel),(x,y))) # and add a (luminance, position) tuple

# Sort the Reference pixels by luminance too
pixelsR.sort()

# Now for the neat observation. Luminance matters more to humans than chromanance,
# given this then we want to match luminance as well as we can. However, we have
# a finite luminance distribution to work with. Since we can't change that, it's best
# just to line the two images up, sorted by luminance, and just simply assign the
# luminance directly. The chrominance will be all kinds of whack, but fixing that
# by way of loose sorting possible chrominance errors takes this algorithm from O(n)
# to O(n^2), which just takes forever (trust me, I've tried it.)

print "Painting reference with palette..."
for p in range(len(pixelsP)): # For each pixel in the palette
    pl,pixel = pixelsP[p] # Grab the pixel from the palette
    l,cord = pixelsR[p] # Grab the location from the reference
    source.putpixel(cord,pixel) # and assign the pallet pixel to the refrence pixels place

print "Applying fixative..."
# save out the result.
source.save("o.png","PNG")

print "Handing it to the master to see if he approves..."
check(palette, source)
print "Done!"

O resultado final tem algumas conseqüências legais. No entanto, se as imagens tiverem distribuições de luminância muito diferentes, não há muito o que fazer sem avançar e hesitar, o que pode ser uma coisa interessante a se fazer em algum momento, mas é excluída aqui por uma questão de brevidade.

Tudo -> Mona Lisa

Gótico Americano -> Mona Lisa Noite Estrelada -> Mona Lisa Gritar -> Mona Lisa Rio -> Mona Lisa Esferas -> Mona Lisa

Mona Lisa -> Esferas

Mona Lisa -> Esferas


6

Mathematica - permutações aleatórias

Idéia

Selecione dois pixels na imagem de origem e verifique se o erro na imagem de destino diminuiria se esses dois pixels fossem trocados. Adicionamos um pequeno número aleatório (-d | + d) ao resultado para evitar mínimos locais. Repetir. Para velocidade, faça isso com 10000 pixels de uma vez.

É um pouco como uma cadeia aleatória de Markov. Provavelmente seria bom diminuir a aleatoriedade durante o processo de otimização semelhante ao recozimento simulado.

Código

colorSpace = "RGB";
\[Delta] = 0.05;
ClearAll[loadImgur, imgToList, listToImg, improveN, err, rearrange, \
rearrangeImg]
loadImgur[tag_] := 
 RemoveAlphaChannel@
  Import["http://i.stack.imgur.com/" <> tag <> ".png"]
imgToList[img_] := Flatten[ImageData[ColorConvert[img, colorSpace]], 1]
listToImg[u_, w_] := Image[Partition[u, w], ColorSpace -> colorSpace]
err[{x_, y_, z_}] := x^2 + y^2 + z^2
improveN[a_, t_, n_] := Block[{i, j, ai, aj, ti, tj},
  {i, j} = Partition[RandomSample[Range@Length@a, 2 n], n];
  ai = a[[i]];
  aj = a[[j]];
  ti = t[[i]];
  tj = t[[j]];
  ReplacePart[
   a, (#1 -> #3) & @@@ 
    Select[Transpose[{i, 
       err /@ (ai - ti) + err /@ (aj - tj) - err /@ (ai - tj) - 
        err /@ (aj - ti) + RandomReal[\[Delta]^2 {-1, +1}, n], aj}], #[[2]] > 0 &]]
  ]
rearrange[ua_, ub_, iterations_: 100] := Block[{tmp = ua},
  Do[tmp = improveN[tmp, ub, Floor[.1 Length@ua]];, {i, iterations}]; 
  tmp]
rearrangeImg[a_, b_, iterations_: 100] := With[{imgdst = loadImgur[b]},
  listToImg[rearrange[
    RandomSample@imgToList@loadImgur[a],
    imgToList@imgdst, iterations], First@ImageDimensions@imgdst]]
rearrangeImg["JXgho","itzIe"]

Resultados

Gótico para Mona Lisa. Esquerda: Usando espaço de cores LAB (delta = 0). Direita: Usando espaço de cores RBG (delta = 0) img7 img8

Gótico para Mona Lisa. Esquerda: espaço de cor RGB, delta = 0,05. Direita: espaço de cores RGB, delta = 0,15. img9 img10

As imagens a seguir mostram animações para cerca de 3.500.000 swaps com espaço de cores RGB e delta = 0.

img1 img2 img3 img4 img5 img6


Parece o caminho do aditsu, mas estou ansioso pelos seus resultados.
Leif

5

Em processamento

A fonte e a paleta são mostradas lado a lado e há uma animação dos pixels tirados da paleta;

Na linha int i = chooseIndexIncremental();, você pode alterar ochooseIndex* funções para ver a ordem de seleção dos pixels.

int scanRate = 20; // pixels per frame

// image filenames
String source = "N6IGO.png";
String palette = "JXgho.png";

PImage src, pal, res;
int area;
int[] lut;
boolean[] processed;
boolean[] taken;
int count = 0;

void start() {
  //size(800, 600);

  src = loadImage(source);
  pal = loadImage(palette);

  size(src.width + pal.width, max(src.height, pal.height));

  src.loadPixels();
  pal.loadPixels();

  int areaSrc = src.pixels.length;
  int areaPal = pal.pixels.length;

  if (areaSrc != areaPal) {
    println("Areas mismatch: src: " + areaSrc + ", pal: " + areaPal);
    return;
  }

  area = areaSrc;

  println("Area: " + area);

  lut = new color[area];
  taken = new boolean[area];
  processed = new boolean[area];

  randomSeed(1);
}

void draw() {
  background(0);
  image(src, 0, 0);
  image(pal, src.width, 0);

  for (int k = 0; k < scanRate; k ++)
  if (count < area) {
    // choose from chooseIndexRandom, chooseIndexSkip and chooseIndexIncremental
    int i = chooseIndexIncremental();
    process(i);

    processed[i] = true;
    count ++;
  }
}

int chooseIndexRandom() {
  int i = 0;
  do i = (int) random(area); while (processed[i]);
  return i;
}

int chooseIndexSkip(int n) {
  int i = (n * count) % area;
  while (processed[i] || i >= area) i = (int) random(area);
  return i;
}

int chooseIndexIncremental() {
  return count;
}

void process(int i) {
  lut[i] = findPixel(src.pixels[i]);
  taken[lut[i]] = true;

  src.loadPixels();
  src.pixels[i] = pal.pixels[lut[i]];
  src.updatePixels();

  pal.loadPixels();
  pal.pixels[lut[i]] = color(0);
  pal.updatePixels();

  stroke(src.pixels[i]);
  int sy = i / src.width;
  int sx = i % src.width;

  int j = lut[i];
  int py = j / pal.width;
  int px = j % pal.width;
  line(sx, sy, src.width + px, py);
}

int findPixel(color c) {
  int best;
  do best = (int) random(area); while (taken[best]);
  float bestDist = colorDist(c, pal.pixels[best]);

  for (int k = 0; k < area; k ++) {
    if (taken[k]) continue;
    color c1 = pal.pixels[k];
    float dist = colorDist(c, c1);
    if (dist < bestDist) {
      bestDist = dist;
      best = k;
    }
  }
  return best;
}

float colorDist(color c1, color c2) {
  return S(red(c1) - red(c2)) + S(green(c1) - green(c2)) + S(blue(c1) - blue(c2));
}

float S(float x) { return x * x; }

Gótico Americano -> Mona Lisa, incremental

incremental

Gótico Americano -> Mona Lisa, aleatória

aleatória


2
Como fica se você usar a paleta Rainbow Spheres?
trate seus mods bem

5

C-Sharp

Nenhuma idéia nova / empolgante, mas pensei em tentar. Simplesmente classifica os pixels, priorizando o brilho sobre a saturação sobre o tom. O código é razoavelmente curto, porém, para o que vale a pena.

EDIT: adicionada uma versão ainda mais curta

using System;
using System.Drawing;
using System.Collections.Generic;

class Program
{
    static void Main(string[] args)
    {
        Bitmap sourceImg = new Bitmap("TheScream.png"),
            arrangImg = new Bitmap("StarryNight.png"),
            destImg = new Bitmap(arrangImg.Width, arrangImg.Height);

        List<Pix> sourcePixels = new List<Pix>(), arrangPixels = new List<Pix>();

        for (int i = 0; i < sourceImg.Width; i++)
            for (int j = 0; j < sourceImg.Height; j++)
                sourcePixels.Add(new Pix(sourceImg.GetPixel(i, j), i, j));

        for (int i = 0; i < arrangImg.Width; i++)
            for (int j = 0; j < arrangImg.Height; j++)
                arrangPixels.Add(new Pix(arrangImg.GetPixel(i, j), i, j));

        sourcePixels.Sort();
        arrangPixels.Sort();

        for (int i = 0; i < arrangPixels.Count; i++)
            destImg.SetPixel(arrangPixels[i].x,
                             arrangPixels[i].y,
                             sourcePixels[i].col);

        destImg.Save("output.png");
    }
}

class Pix : IComparable<Pix>
{
    public Color col;
    public int x, y;
    public Pix(Color col, int x, int y)
    {
        this.col = col;
        this.x = x;
        this.y = y;
    }

    public int CompareTo(Pix other)
    {
        return(int)(255 * 255 * 255 * (col.GetBrightness() - other.col.GetBrightness())
                + (255 * (col.GetHue() - other.col.GetHue()))
                + (255 * 255 * (col.GetSaturation() - other.col.GetSaturation())));
    }
}

Amostra:

insira a descrição da imagem aqui

+

insira a descrição da imagem aqui

=

insira a descrição da imagem aqui


5

Java

import java.awt.image.BufferedImage;
import java.io.File;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Collections;
import java.util.Iterator;
import java.util.List;
import javax.imageio.ImageIO;

/**
 *
 * @author Quincunx
 */
public class PixelRearrangerMK2 {

    public static void main(String[] args) throws IOException {
        BufferedImage source = ImageIO.read(resource("Raytraced Spheres.png"));
        BufferedImage palette = ImageIO.read(resource("American Gothic.png"));
        BufferedImage result = rearrange(source, palette);
        ImageIO.write(result, "png", resource("result.png"));
        validate(palette, result);
    }

    public static BufferedImage rearrange(BufferedImage source, BufferedImage palette) {
        List<Color> sColors = Color.getColors(source);
        List<Color> pColors = Color.getColors(palette);
        Collections.sort(sColors);
        Collections.sort(pColors);

        BufferedImage result = new BufferedImage(source.getWidth(), source.getHeight(), BufferedImage.TYPE_INT_RGB);
        Iterator<Color> sIter = sColors.iterator();
        Iterator<Color> pIter = pColors.iterator();

        while (sIter.hasNext()) {
            Color s = sIter.next();
            Color p = pIter.next();

            result.setRGB(s.x, s.y, p.rgb);
        }
        return result;
    }

    public static class Color implements Comparable {
        int x, y;
        int rgb;
        double hue;

        private int r, g, b;

        public Color(int x, int y, int rgb) {
            this.x = x;
            this.y = y;
            this.rgb = rgb;
            r = (rgb & 0xFF0000) >> 16;
            g = (rgb & 0x00FF00) >> 8;
            b = rgb & 0x0000FF;
            hue = Math.atan2(Math.sqrt(3) * (g - b), 2 * r - g - b);
        }

        @Override
        public int compareTo(Object o) {
            Color c = (Color) o;
            return hue < c.hue ? -1 : hue == c.hue ? 0 : 1;
        }

        public static List<Color> getColors(BufferedImage img) {
            List<Color> result = new ArrayList<>();
            for (int y = 0; y < img.getHeight(); y++) {
                for (int x = 0; x < img.getWidth(); x++) {
                    result.add(new Color(x, y, img.getRGB(x, y)));
                }
            }
            return result;
        }
    }

    //Validation and util methods follow
    public static void validate(BufferedImage palette, BufferedImage result) {
        List<Integer> paletteColors = getColorsAsInt(palette);
        List<Integer> resultColors = getColorsAsInt(result);
        Collections.sort(paletteColors);
        Collections.sort(resultColors);
        System.out.println(paletteColors.equals(resultColors));
    }

    public static List<Integer> getColorsAsInt(BufferedImage img) {
        int width = img.getWidth();
        int height = img.getHeight();
        List<Integer> colors = new ArrayList<>(width * height);
        for (int x = 0; x < width; x++) {
            for (int y = 0; y < height; y++) {
                colors.add(img.getRGB(x, y));
            }
        }
        Collections.sort(colors);
        return colors;
    }

    public static File resource(String fileName) {
        return new File(System.getProperty("user.dir") + "/Resources/" + fileName);
    }
}

Aqui está uma ideia completamente diferente. Crio uma lista das cores de cada imagem e, em seguida, classifico de acordo com a tonalidade, calculada pela fórmula da wikipedia:

insira a descrição da imagem aqui

Ao contrário da minha outra resposta, isso é extremamente rápido. Demora cerca de 2 segundos, incluindo a validação.

O resultado é uma arte abstrata. Aqui estão algumas imagens (passe o mouse para ver de / para):

insira a descrição da imagem aqui insira a descrição da imagem aqui insira a descrição da imagem aqui insira a descrição da imagem aqui insira a descrição da imagem aqui insira a descrição da imagem aqui insira a descrição da imagem aqui


5
Parece que algo que o Predator veria o_O
publicado em 11/07/2014

Estes são bastante assustadores, mas são realmente corretos!
9789 Calvin's Hobbies

1
@ Calvin'sHobbies Como isso é assustador? Eu chamo isso de beleza.
11117 Justin

3
Seus rostos estão vazios e misteriosos ... mas eles têm uma beleza assustadora.
9789 Calvin's Hobbies

1
As esferas são incríveis.
siledh

5

Pitão

Bem, decidi que também poderia postar minha solução, pois gastei tempo para fazê-lo. Basicamente, o que imaginei que faria seria obter os dados brutos de pixel das imagens, classificar os dados por brilho e, em seguida, colocar os valores do mesmo índice em uma nova imagem. Mudei de idéia sobre o brilho e usei a luminância. Eu obtive alguns bons resultados com isso.

from PIL import Image
from optparse import OptionParser


def key_func(arr):
    # Sort the pixels by luminance
    r = 0.2126*arr[0] + 0.7152*arr[1] + 0.0722*arr[2]
    return r


def main():
    # Parse options from the command line
    parser = OptionParser()
    parser.add_option("-p", "--pixels", dest="pixels",
                      help="use pixels from FILE", metavar="FILE")
    parser.add_option("-i", "--input", dest="input", metavar="FILE",
                      help="recreate FILE")
    parser.add_option("-o", "--out", dest="output", metavar="FILE",
                      help="output to FILE", default="output.png")

    (options, args) = parser.parse_args()

    if not options.pixels or not options.input:
        raise Exception("Missing arguments. See help for more info.")

    # Load the images
    im1 = Image.open(options.pixels)
    im2 = Image.open(options.input)

    # Get the images into lists
    px1 = list(im1.getdata())
    px2 = list(im2.getdata())
    w1, h1 = im1.size
    w2, h2 = im2.size

    if w1*h1 != w2*h2:
        raise Exception("Images must have the same number of pixels.")

    # Sort the pixels lists by luminance
    px1_s = sorted(px1, key=key_func)
    px2_s = sorted(px2, key=key_func)

    # Create an array of nothing but black pixels
    arr = [(0, 0, 0)]*w2*h2

    # Create a dict that contains a list of locations with pixel value as key
    # This speeds up the process a lot, since before it was O(n^2)
    locations_cache = {}
    for index, val in enumerate(px2):
        v = str(val)
        if v in locations_cache:
            locations_cache[v].append(index)
        else:
            locations_cache[v] = [index]

    # Loop through each value of the sorted pixels
    for index, val in enumerate(px2_s):
        # Find the original location of the pixel
        # v = px2.index(val)
        v = locations_cache[str(val)].pop(0)
        # Set the value of the array at the given location to the pixel of the
        # equivalent luminance from the source image
        arr[v] = px1_s[index]
        # v2 = px1.index(px1_s[index])
        # Set the value of px2 to an arbitrary value outside of the RGB range
        # This prevents duplicate pixel locations
        # I would use "del px2[v]", but it wouldn't work for some reason
        px2[v] = (512, 512, 512)
        # px1[v2] = (512, 512, 512)
        # Print the percent progress
        print("%f%%" % (index/len(px2)*100))
        """if index % 500 == 0 or index == len(px2_s)-1:
            if h1 > h2:
                size = (w1+w2, h1)
            else:
                size = (w1+w2, h2)
            temp_im1 = Image.new("RGB", im2.size)
            temp_im1.putdata(arr)

            temp_im2 = Image.new("RGB", im1.size)
            temp_im2.putdata(px1)

            temp_im3 = Image.new("RGB", size)
            temp_im3.paste(temp_im1, (0, 0))
            temp_im3.paste(temp_im2, (w2, 0))
            temp_im3.save("still_frames/img_%04d.png" % (index/500))"""

    # Save the image
    im3 = Image.new('RGB', im2.size)
    im3.putdata(arr)
    im3.save(options.output)

if __name__ == '__main__':
    main()

Resultados

Fiquei muito feliz com os resultados. Parecia funcionar de forma consistente para todas as imagens que eu coloquei nele.

Noite estrelada com pixels de grito

Grito + Noite Estrelada

Noite estrelada com pixels do arco-íris

Arco-íris + noite estrelada

Arco-íris com pixels da noite estrelada

Noite estrelada + arco-íris

Mona Lisa com Scream Pixels

Grito + Mona Lisa

Rio com pixels da noite estrelada

Noite estrelada + rio

Mona Lisa com Pixels Góticos Americanos

Gótico + Mona Lisa

Mustang com Chevy Pixels

Eu provavelmente deveria ter reduzido as imagens devido às minhas restrições de hardware, mas tudo bem.

Chevy + Mustang

Chevy com Pixels Mustang

Mustang + Chevy

Rio com pixels do arco-íris

Arco-íris + Rio

Mona Lisa com pixels do arco-íris

Arco-íris + Mona Lisa

Gótico americano com pixels do arco-íris

Rainbow + Gothic


Atualização Adicionei mais algumas fotos e aqui estão algumas animações. O primeiro mostra como meu método funcionou e o segundo está usando o script @ Calvin'sHobbies publicado.

Meu método

@ Calvin'sHobbies script


Atualização 2 Adicionei um ditado armazenando os índices de pixels por sua cor. Isso tornou o script muito mais eficiente. Para ver o original, verifique o histórico de revisões.


5

C ++ 11

No final, decidi por um algoritmo ganancioso determinístico relativamente simples. Esta é uma rosca única, mas corre em um fio de cabelo durante 4 segundos na minha máquina.

O algoritmo básico funciona classificando todos os pixels na paleta e na imagem de destino diminuindo a luminância (o L de L a b * ). Em seguida, para cada um desses pixels de destino ordenados, ele procura a correspondência mais próxima nas primeiras 75 entradas da paleta, usando o quadrado da métrica de distância CIEDE2000 com a luminância das cores da paleta presas à do alvo. (Para implementação e depuração do CIEDE2000, esta página foi muito útil). A melhor correspondência é removida da paleta, atribuída ao resultado e o algoritmo passa para o próximo pixel mais claro na imagem de destino.

Usando a luminância classificada para o alvo e a paleta, garantimos que a luminância geral (o elemento mais saliente visualmente) do resultado corresponda ao alvo o mais próximo possível. O uso de uma pequena janela de 75 entradas fornece uma boa chance de encontrar uma cor correspondente ao brilho certo, se houver. Se não houver, a cor será desativada, mas pelo menos o brilho deverá ser consistente. Como resultado, ela se degrada bastante quando as cores da paleta não combinam bem.

Código

Para compilar isso, você precisará das bibliotecas de desenvolvimento ImageMagick ++. Um pequeno arquivo CMake para compilá-lo também está incluído abaixo.

palette.cpp

#include <Magick++.h>
#include <algorithm>
#include <functional>
#include <utility>
#include <set>

using namespace std;
using namespace Magick;

struct Lab
{
    PixelPacket rgb;
    float L, a, b;

    explicit Lab(
        PixelPacket rgb )
        : rgb( rgb )
    {
        auto R_srgb = static_cast< float >( rgb.red ) / QuantumRange;
        auto G_srgb = static_cast< float >( rgb.green ) / QuantumRange;
        auto B_srgb = static_cast< float >( rgb.blue ) / QuantumRange;
        auto R_lin = R_srgb < 0.04045f ? R_srgb / 12.92f :
            powf( ( R_srgb + 0.055f ) / 1.055f, 2.4f );
        auto G_lin = G_srgb < 0.04045f ? G_srgb / 12.92f :
            powf( ( G_srgb + 0.055f ) / 1.055f, 2.4f );
        auto B_lin = B_srgb < 0.04045f ? B_srgb / 12.92f :
            powf( ( B_srgb + 0.055f ) / 1.055f, 2.4f );
        auto X = 0.4124f * R_lin + 0.3576f * G_lin + 0.1805f * B_lin;
        auto Y = 0.2126f * R_lin + 0.7152f * G_lin + 0.0722f * B_lin;
        auto Z = 0.0193f * R_lin + 0.1192f * G_lin + 0.9502f * B_lin;
        auto X_norm = X / 0.9505f;
        auto Y_norm = Y / 1.0000f;
        auto Z_norm = Z / 1.0890f;
        auto fX = ( X_norm > 216.0f / 24389.0f ?
                    powf( X_norm, 1.0f / 3.0f ) :
                    X_norm * ( 841.0f / 108.0f ) + 4.0f / 29.0f );
        auto fY = ( Y_norm > 216.0f / 24389.0f ?
                    powf( Y_norm, 1.0f / 3.0f ) :
                    Y_norm * ( 841.0f / 108.0f ) + 4.0f / 29.0f );
        auto fZ = ( Z_norm > 216.0f / 24389.0f ?
                    powf( Z_norm, 1.0f / 3.0f ) :
                    Z_norm * ( 841.0f / 108.0f ) + 4.0f / 29.0f );
        L = 116.0f * fY - 16.0f;
        a = 500.0f * ( fX - fY );
        b = 200.0f * ( fY - fZ );
    }

    bool operator<(
        Lab const that ) const
    {
        return ( L > that.L ? true :
                 L < that.L ? false :
                 a > that.a ? true :
                 a < that.a ? false :
                 b > that.b );
    }

    Lab clampL(
        Lab const that ) const
    {
        auto result = Lab( *this );
        if ( result.L > that.L )
            result.L = that.L;
        return result;
    }

    float cieDe2000(
        Lab const that,
        float const k_L = 1.0f,
        float const k_C = 1.0f,
        float const k_H = 1.0f ) const
    {
        auto square = []( float value ){ return value * value; };
        auto degs = []( float rad ){ return rad * 180.0f / 3.14159265359f; };
        auto rads = []( float deg ){ return deg * 3.14159265359f / 180.0f; };
        auto C_1 = hypot( a, b );
        auto C_2 = hypot( that.a, that.b );
        auto C_bar = ( C_1 + C_2 ) * 0.5f;
        auto C_bar_7th = square( square( C_bar ) ) * square( C_bar ) * C_bar;
        auto G = 0.5f * ( 1.0f - sqrtf( C_bar_7th / ( C_bar_7th + 610351562.0f ) ) );
        auto a_1_prime = ( 1.0f + G ) * a;
        auto a_2_prime = ( 1.0f + G ) * that.a;
        auto C_1_prime = hypot( a_1_prime, b );
        auto C_2_prime = hypot( a_2_prime, that.b );
        auto h_1_prime = C_1_prime == 0.0f ? 0.0f : degs( atan2f( b, a_1_prime ) );
        if ( h_1_prime < 0.0f )
            h_1_prime += 360.0f;
        auto h_2_prime = C_2_prime == 0.0f ? 0.0f : degs( atan2f( that.b, a_2_prime ) );
        if ( h_2_prime < 0.0f )
            h_2_prime += 360.0f;
        auto delta_L_prime = that.L - L;
        auto delta_C_prime = C_2_prime - C_1_prime;
        auto delta_h_prime =
            C_1_prime * C_2_prime == 0.0f ? 0 :
            fabs( h_2_prime - h_1_prime ) <= 180.0f ? h_2_prime - h_1_prime :
            h_2_prime - h_1_prime > 180.0f ? h_2_prime - h_1_prime - 360.0f :
            h_2_prime - h_1_prime + 360.0f;
        auto delta_H_prime = 2.0f * sqrtf( C_1_prime * C_2_prime ) *
            sinf( rads( delta_h_prime * 0.5f ) );
        auto L_bar_prime = ( L + that.L ) * 0.5f;
        auto C_bar_prime = ( C_1_prime + C_2_prime ) * 0.5f;
        auto h_bar_prime =
            C_1_prime * C_2_prime == 0.0f ? h_1_prime + h_2_prime :
            fabs( h_1_prime - h_2_prime ) <= 180.0f ? ( h_1_prime + h_2_prime ) * 0.5f :
            h_1_prime + h_2_prime < 360.0f ? ( h_1_prime + h_2_prime + 360.0f ) * 0.5f :
            ( h_1_prime + h_2_prime - 360.0f ) * 0.5f;
        auto T = ( 1.0f
                   - 0.17f * cosf( rads( h_bar_prime - 30.0f ) )
                   + 0.24f * cosf( rads( 2.0f * h_bar_prime ) )
                   + 0.32f * cosf( rads( 3.0f * h_bar_prime + 6.0f ) )
                   - 0.20f * cosf( rads( 4.0f * h_bar_prime - 63.0f ) ) );
        auto delta_theta = 30.0f * expf( -square( ( h_bar_prime - 275.0f ) / 25.0f ) );
        auto C_bar_prime_7th = square( square( C_bar_prime ) ) *
            square( C_bar_prime ) * C_bar_prime;
        auto R_C = 2.0f * sqrtf( C_bar_prime_7th / ( C_bar_prime_7th + 610351562.0f ) );
        auto S_L = 1.0f + ( 0.015f * square( L_bar_prime - 50.0f ) /
                            sqrtf( 20.0f + square( L_bar_prime - 50.0f ) ) );
        auto S_C = 1.0f + 0.045f * C_bar_prime;
        auto S_H = 1.0f + 0.015f * C_bar_prime * T;
        auto R_T = -sinf( rads( 2.0f * delta_theta ) ) * R_C;
        return (
            square( delta_L_prime / ( k_L * S_L ) ) +
            square( delta_C_prime / ( k_C * S_C ) ) +
            square( delta_H_prime / ( k_H * S_H ) ) +
            R_T * delta_C_prime * delta_H_prime / ( k_C * S_C * k_H * S_H ) );
    }

};

Image read_image(
    char * const filename )
{
    auto result = Image( filename );
    result.type( TrueColorType );
    result.matte( true );
    result.backgroundColor( Color( 0, 0, 0, QuantumRange ) );
    return result;
}

template< typename T >
multiset< T > map_image(
    Image const &image,
    function< T( unsigned, PixelPacket ) > const transform )
{
    auto width = image.size().width();
    auto height = image.size().height();
    auto result = multiset< T >();
    auto pixels = image.getConstPixels( 0, 0, width, height );
    for ( auto index = 0; index < width * height; ++index, ++pixels )
        result.emplace( transform( index, *pixels ) );
    return result;
}

int main(
    int argc,
    char **argv )
{
    auto palette = map_image(
        read_image( argv[ 1 ] ),
        function< Lab( unsigned, PixelPacket ) >(
            []( unsigned index, PixelPacket rgb ) {
                return Lab( rgb );
            } ) );

    auto target_image = read_image( argv[ 2 ] );
    auto target_colors = map_image(
        target_image,
        function< pair< Lab, unsigned >( unsigned, PixelPacket ) >(
            []( unsigned index, PixelPacket rgb ) {
                return make_pair( Lab( rgb ), index );
            } ) );

    auto pixels = target_image.setPixels(
        0, 0,
        target_image.size().width(),
        target_image.size().height() );
    for ( auto &&target : target_colors )
    {
        auto best_color = palette.begin();
        auto best_difference = 1.0e38f;
        auto count = 0;
        for ( auto candidate = palette.begin();
              candidate != palette.end() && count < 75;
              ++candidate, ++count )
        {
            auto difference = target.first.cieDe2000(
                candidate->clampL( target.first ) );
            if ( difference < best_difference )
            {
                best_color = candidate;
                best_difference = difference;
            }
        }
        pixels[ target.second ] = best_color->rgb;
        palette.erase( best_color );
    }
    target_image.syncPixels();
    target_image.write( argv[ 3 ] );

    return 0;
}

CMakeList.txt

cmake_minimum_required( VERSION 2.8.11 )
project( palette )
add_executable( palette palette.cpp)
find_package( ImageMagick COMPONENTS Magick++ )
if( ImageMagick_FOUND )
    include_directories( ${ImageMagick_INCLUDE_DIRS} )
    target_link_libraries( palette ${ImageMagick_LIBRARIES} )
endif( ImageMagick_FOUND )

Resultado

O álbum completo está aqui. Dos resultados abaixo, meus favoritos são provavelmente American Gothic com a paleta Mona Lisa e Starry Night com a paleta Spheres.

Paleta gótica americana

Mona Lisa Palette

River Palette

The Scream Palette

Paleta Esferas

Starry Night Palette


Isso parece fantástico! O que você pensa sobre o quanto isso pode ser acelerado? Existe uma chance em tempo real, ou seja, 60fps em hardware médio?
danijar

4

C ++

Não é o código mais curto, mas gera a resposta 'instantaneamente', apesar de ser de thread único e não otimizado. Eu estou satisfeito com os resultados.

Gero duas listas classificadas de pixels, uma para cada imagem, e a classificação é baseada em um valor ponderado de 'brilho'. Uso 100% verde, 50% vermelho e 10% azul para calcular o brilho, ponderando-o para o olho humano (mais ou menos). Troco pixels na imagem de origem pelo mesmo pixel indexado na imagem da paleta e escrevo a imagem de destino.

Eu uso a biblioteca FreeImage para ler / gravar os arquivos de imagem.

Código

/* Inputs: 2 image files of same area
Outputs: image1 made from pixels of image2*/
#include <iostream>
#include <stdlib.h>
#include "FreeImage.h"
#include <vector>
#include <algorithm>

class pixel
{
public:
    int x, y;
    BYTE r, g, b;
    float val;  //color value; weighted 'brightness'
};

bool sortByColorVal(const pixel &lhs, const pixel &rhs) { return lhs.val > rhs.val; }

FIBITMAP* GenericLoader(const char* lpszPathName, int flag) 
{
    FREE_IMAGE_FORMAT fif = FIF_UNKNOWN;

    // check the file signature and deduce its format
    // (the second argument is currently not used by FreeImage)
    fif = FreeImage_GetFileType(lpszPathName, 0);
    if (fif == FIF_UNKNOWN) 
    {
        // no signature ?
        // try to guess the file format from the file extension
        fif = FreeImage_GetFIFFromFilename(lpszPathName);
    }
    // check that the plugin has reading capabilities ...
    if ((fif != FIF_UNKNOWN) && FreeImage_FIFSupportsReading(fif)) 
    {
        // ok, let's load the file
        FIBITMAP *dib = FreeImage_Load(fif, lpszPathName, flag);
        // unless a bad file format, we are done !
        return dib;
    }
    return NULL;
}

bool GenericWriter(FIBITMAP* dib, const char* lpszPathName, int flag) 
{
    FREE_IMAGE_FORMAT fif = FIF_UNKNOWN;
    BOOL bSuccess = FALSE;

    if (dib) 
    {
        // try to guess the file format from the file extension
        fif = FreeImage_GetFIFFromFilename(lpszPathName);
        if (fif != FIF_UNKNOWN) 
        {
            // check that the plugin has sufficient writing and export capabilities ...
            WORD bpp = FreeImage_GetBPP(dib);
            if (FreeImage_FIFSupportsWriting(fif) && FreeImage_FIFSupportsExportBPP(fif, bpp)) 
            {
                // ok, we can save the file
                bSuccess = FreeImage_Save(fif, dib, lpszPathName, flag);
                // unless an abnormal bug, we are done !
            }
        }
    }
    return (bSuccess == TRUE) ? true : false;
}

void FreeImageErrorHandler(FREE_IMAGE_FORMAT fif, const char *message) 
{
    std::cout << std::endl << "*** ";
    if (fif != FIF_UNKNOWN) 
    {
        std::cout << "ERROR: " << FreeImage_GetFormatFromFIF(fif) << " Format" << std::endl;
    }
    std::cout << message;
    std::cout << " ***" << std::endl;
}

FIBITMAP* Convert24BPP(FIBITMAP* dib)
{
    if (FreeImage_GetBPP(dib) == 24) return dib;

    FIBITMAP *dib2 = FreeImage_ConvertTo24Bits(dib);
    FreeImage_Unload(dib);
    return dib2;
}
// ----------------------------------------------------------

int main(int argc, char **argv)
{
    // call this ONLY when linking with FreeImage as a static library
#ifdef FREEIMAGE_LIB
    FreeImage_Initialise();
#endif

    FIBITMAP *src = NULL, *pal = NULL;
    int result = EXIT_FAILURE;

    // initialize my own FreeImage error handler
    FreeImage_SetOutputMessage(FreeImageErrorHandler);

    // print version
    std::cout << "FreeImage version : " << FreeImage_GetVersion() << std::endl;

    if (argc != 4) 
    {
        std::cout << "USAGE : Pic2Pic <source image> <palette image> <output file name>" << std::endl;
        return EXIT_FAILURE;
    }

    // Load the src image
    src = GenericLoader(argv[1], 0);
    if (src) 
    {
        // load the palette image
        pal = GenericLoader(argv[2], 0);

        if (pal) 
        {
            //compare areas
            // if(!samearea) return EXIT_FAILURE;
            unsigned int width_src = FreeImage_GetWidth(src);
            unsigned int height_src = FreeImage_GetHeight(src);
            unsigned int width_pal = FreeImage_GetWidth(pal);
            unsigned int height_pal = FreeImage_GetHeight(pal);

            if (width_src * height_src != width_pal * height_pal)
            {
                std::cout << "ERROR: source and palette images do not have the same pixel area." << std::endl;
                result = EXIT_FAILURE;
            }
            else
            {
                //go to work!

                //first make sure everything is 24 bit:
                src = Convert24BPP(src);
                pal = Convert24BPP(pal);

                //retrieve the image data
                BYTE *bits_src = FreeImage_GetBits(src);
                BYTE *bits_pal = FreeImage_GetBits(pal);

                //make destination image
                FIBITMAP *dst = FreeImage_ConvertTo24Bits(src);
                BYTE *bits_dst = FreeImage_GetBits(dst);

                //make a vector of all the src pixels that we can sort by color value
                std::vector<pixel> src_pixels;
                for (unsigned int y = 0; y < height_src; ++y)
                {
                    for (unsigned int x = 0; x < width_src; ++x)
                    {
                        pixel p;
                        p.x = x;
                        p.y = y;

                        p.b = bits_src[y*width_src * 3 + x * 3];
                        p.g = bits_src[y*width_src * 3 + x * 3 + 1];
                        p.r = bits_src[y*width_src * 3 + x * 3 + 2];

                        //calculate color value using a weighted brightness for each channel
                        //p.val = 0.2126f * p.r + 0.7152f * p.g + 0.0722f * p.b; //from http://www.poynton.com/notes/colour_and_gamma/ColorFAQ.html
                        p.val = 0.5f * p.r + p.g + 0.1f * p.b;                      

                        src_pixels.push_back(p);
                    }
                }

                //sort by color value
                std::sort(src_pixels.begin(), src_pixels.end(), sortByColorVal);

                //make a vector of all palette pixels we can use
                std::vector<pixel> pal_pixels;

                for (unsigned int y = 0; y < height_pal; ++y)
                {
                    for (unsigned int x = 0; x < width_pal; ++x)
                    {
                        pixel p;

                        p.b = bits_pal[y*width_pal * 3 + x * 3];
                        p.g = bits_pal[y*width_pal * 3 + x * 3 + 1];
                        p.r = bits_pal[y*width_pal * 3 + x * 3 + 2];

                        p.val = 0.5f * p.r + p.g + 0.1f * p.b;

                        pal_pixels.push_back(p);
                    }
                }

                //sort by color value
                std::sort(pal_pixels.begin(), pal_pixels.end(), sortByColorVal);

                //for each src pixel, match it with same index palette pixel and copy to destination
                for (unsigned int i = 0; i < width_src * height_src; ++i)
                {
                    bits_dst[src_pixels[i].y * width_src * 3 + src_pixels[i].x * 3] = pal_pixels[i].b;
                    bits_dst[src_pixels[i].y * width_src * 3 + src_pixels[i].x * 3 + 1] = pal_pixels[i].g;
                    bits_dst[src_pixels[i].y * width_src * 3 + src_pixels[i].x * 3 + 2] = pal_pixels[i].r;
                }

                // Save the destination image
                bool bSuccess = GenericWriter(dst, argv[3], 0);
                if (!bSuccess)
                {
                    std::cout << "ERROR: unable to save " << argv[3] << std::endl;
                    std::cout << "This format does not support 24-bit images" << std::endl;
                    result = EXIT_FAILURE;
                }
                else result = EXIT_SUCCESS;

                FreeImage_Unload(dst);
            }

            // Free pal
            FreeImage_Unload(pal);
        }

        // Free src
        FreeImage_Unload(src);
    }

#ifdef FREEIMAGE_LIB
    FreeImage_DeInitialise();
#endif

    if (result == EXIT_SUCCESS) std::cout << "SUCCESS!" << std::endl;
    else std::cout << "FAILURE!" << std::endl;
    return result;
}

Resultados

Gótico americano usando a paleta Mona Lisa American Gothic usando a paleta Mona Lisa Gótico americano usando a paleta Rainbow American Gothic usando a paleta Rainbow Mona Lisa usando a paleta Scream Mona Lisa usando a paleta Scream Mona Lisa usando a paleta Rainbow Mona Lisa usando a paleta Rainbow Gritar usando a paleta Noite Estrelada Scream usando a paleta Starry Night


4

C #

Os pontos são ordenados em caminhada aleatória, a partir do centro. obtenha sempre a cor mais próxima da imagem da paleta. Portanto, os últimos pixels são um pouco muito ruins.

Resultados

Paleta Gótica

insira a descrição da imagem aqui

insira a descrição da imagem aqui

E o casal de visitantes americanos da wikipedia

insira a descrição da imagem aqui

Mona Palette

insira a descrição da imagem aqui

insira a descrição da imagem aqui

insira a descrição da imagem aqui

Código:

Não sei por que, mas o código é bem lento ...

public class PixelExchanger
{
    public class ProgressInfo
    {
        public readonly Pixel NewPixel;
        public readonly int Percentage;

        public ProgressInfo(Pixel newPixel, int percentage)
        {
            this.NewPixel = newPixel;
            this.Percentage = percentage;
        }
    }

    public class Pixel
    {
        public readonly int X;
        public readonly int Y;
        public readonly Color Color;

        public Pixel(int x, int y, Color color)
        {
            this.X = x;
            this.Y = y;
            this.Color = color;
        }
    }

    private static Random r = new Random(0);

    private readonly Bitmap Pallete;
    private readonly Bitmap Image;

    private readonly int Width;
    private readonly int Height;

    private readonly Action<ProgressInfo> ProgressCallback;
    private System.Drawing.Image image1;
    private System.Drawing.Image image2;

    private int Area { get { return Width * Height; } }

    public PixelExchanger(Bitmap pallete, Bitmap image, Action<ProgressInfo> progressCallback = null)
    {
        this.Pallete = pallete;
        this.Image = image;

        this.ProgressCallback = progressCallback;

        Width = image.Width;
        Height = image.Height;

        if (Area != pallete.Width * pallete.Height)
            throw new ArgumentException("Image and Pallete have different areas!");
    }

    public Bitmap DoWork()
    {
        var array = GetColorArray();
        var map = GetColorMap(Image);
        var newMap = Go(array, map);

        var bm = new Bitmap(map.Length, map[0].Length);

        for (int i = 0; i < Width; i++)
        {
            for (int j = 0; j < Height; j++)
            {
                bm.SetPixel(i, j, newMap[i][j]);
            }
        }

        return bm;
    }

    public Color[][] Go(List<Color> array, Color[][] map)
    {
        var centralPoint = new Point(Width / 2, Height / 2);

        var q = OrderRandomWalking(centralPoint).ToArray();

        Color[][] newMap = new Color[map.Length][];
        for (int i = 0; i < map.Length; i++)
        {
            newMap[i] = new Color[map[i].Length];
        }

        double pointsDone = 0;

        foreach (var p in q)
        {
            newMap[p.X][p.Y] = Closest(array, map[p.X][p.Y]);

            pointsDone++;

            if (ProgressCallback != null)
            {
                var percent = 100 * (pointsDone / (double)Area);

                var progressInfo = new ProgressInfo(new Pixel(p.X, p.Y, newMap[p.X][p.Y]), (int)percent);

                ProgressCallback(progressInfo);
            }
        }

        return newMap;
    }

    private int[][] GetCardinals()
    {
        int[] nn = new int[] { -1, +0 };
        // int[] ne = new int[] { -1, +1 };
        int[] ee = new int[] { +0, +1 };
        // int[] se = new int[] { +1, +1 };
        int[] ss = new int[] { +1, +0 };
        // int[] sw = new int[] { +1, -1 };
        int[] ww = new int[] { +0, -1 };
        // int[] nw = new int[] { -1, -1 };

        var dirs = new List<int[]>();

        dirs.Add(nn);
        // dirs.Add(ne);
        dirs.Add(ee);
        // dirs.Add(se);
        dirs.Add(ss);
        // dirs.Add(sw);
        dirs.Add(ww);
        // dirs.Add(nw);

        return dirs.ToArray();
    }

    private Color Closest(List<Color> array, Color c)
    {
        int closestIndex = -1;

        int bestD = int.MaxValue;

        int[] ds = new int[array.Count];
        Parallel.For(0, array.Count, (i, state) =>
        {
            ds[i] = Distance(array[i], c);

            if (ds[i] <= 50)
            {
                closestIndex = i;
                state.Break();
            }
            else if (bestD > ds[i])
            {
                bestD = ds[i];
                closestIndex = i;
            }
        });

        var closestColor = array[closestIndex];

        array.RemoveAt(closestIndex);

        return closestColor;
    }

    private int Distance(Color c1, Color c2)
    {
        var r = Math.Abs(c1.R - c2.R);
        var g = Math.Abs(c1.G - c2.G);
        var b = Math.Abs(c1.B - c2.B);
        var s = Math.Abs(c1.GetSaturation() - c1.GetSaturation());

        return (int)s + r + g + b;
    }

    private HashSet<Point> OrderRandomWalking(Point p)
    {
        var points = new HashSet<Point>();

        var dirs = GetCardinals();
        var dir = new int[] { 0, 0 };

        while (points.Count < Width * Height)
        {
            bool inWidthBound = p.X + dir[0] < Width && p.X + dir[0] >= 0;
            bool inHeightBound = p.Y + dir[1] < Height && p.Y + dir[1] >= 0;

            if (inWidthBound && inHeightBound)
            {
                p.X += dir[0];
                p.Y += dir[1];

                points.Add(p);
            }

            dir = dirs.Random(r);
        }

        return points;
    }

    private static Color[][] GetColorMap(Bitmap b1)
    {
        int hight = b1.Height;
        int width = b1.Width;

        Color[][] colorMatrix = new Color[width][];
        for (int i = 0; i < width; i++)
        {
            colorMatrix[i] = new Color[hight];
            for (int j = 0; j < hight; j++)
            {
                colorMatrix[i][j] = b1.GetPixel(i, j);
            }
        }
        return colorMatrix;
    }

    private List<Color> GetColorArray()
    {
        var map = GetColorMap(Pallete);

        List<Color> colors = new List<Color>();

        foreach (var line in map)
        {
            colors.AddRange(line);
        }

        return colors;
    }
}

2
Estes são ótimos. Eles parecem fotos que foram queimadas ou deixadas em algum lugar para apodrecer.

Obrigado, A fez vários algoritmos, mas o outro era muito parecido com as outras respostas. Então, eu postei o mais distintivo
RMalke

3

C #

Compara as cores pela distância que estão. Começa a partir do centro.

EDIT: atualizado, agora deve ser cerca de 1,5x mais rápido.

American Gothic
insira a descrição da imagem aqui
The Scream
insira a descrição da imagem aqui
Starry Night
insira a descrição da imagem aqui
Marbles
insira a descrição da imagem aqui
River
insira a descrição da imagem aqui
Além disso, aqui está o Chevy amarelo:
insira a descrição da imagem aqui

using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Drawing;
using System.Threading.Tasks;
using System.Diagnostics;

namespace ConsoleApplication1
{
    class Pixel
    {
        public int X = 0;
        public int Y = 0;
        public Color Color = new Color();
        public Pixel(int x, int y, Color clr)
        {
            Color = clr;
            X = x;
            Y = y;
        }
        public Pixel()
        {
        }
    }
    class Vector2
    {
        public int X = 0;
        public int Y = 0;
        public Vector2(int x, int y)
        {
            X = x;
            Y = y;
        }
        public Vector2()
        {
        }
        public double Diagonal()
        {
            return Math.Sqrt((X * X) + (Y * Y));
        }
    }
    class ColorCollection
    {
        Dictionary<Color, int> dict = new Dictionary<Color, int>();
        public ColorCollection()
        {
        }
        public void AddColor(Color color)
        {
            if (dict.ContainsKey(color))
            {
                dict[color]++;
                return;
            }
            dict.Add(color, 1);
        }
        public void UseColor(Color color)
        {
            if (dict.ContainsKey(color))
                dict[color]--;
            if (dict[color] < 1)
                dict.Remove(color);
        }
        public Color FindBestColor(Color color)
        {
            Color ret = dict.First().Key;
            int p = this.CalculateDifference(ret, color);
            foreach (KeyValuePair<Color, int> pair in dict)
            {
                int points = CalculateDifference(pair.Key, color);
                if (points < p)
                {
                    ret = pair.Key;
                    p = points;
                }
            }
            this.UseColor(ret);
            return ret;
        }
        int CalculateDifference(Color c1, Color c2)
        {
            int ret = 0;
            ret = ret + Math.Abs(c1.R - c2.R);
            ret = ret + Math.Abs(c1.G - c2.G);
            ret = ret + Math.Abs(c1.B - c2.B);
            return ret;
        }
    }

    class Program
    {
        static void Main(string[] args)
        {
            string img1 = "";
            string img2 = "";
            if (args.Length != 2)
            {
                Console.Write("Where is the first picture located? ");
                img1 = Console.ReadLine();
                Console.Write("Where is the second picture located? ");
                img2 = Console.ReadLine();
            }
            else
            {
                img1 = args[0];
                img2 = args[1];
            }
            Bitmap bmp1 = new Bitmap(img1);
            Bitmap bmp2 = new Bitmap(img2);
            Console.WriteLine("Getting colors....");
            ColorCollection colors = GetColors(bmp1);
            Console.WriteLine("Getting pixels....");
            List<Pixel> pixels = GetPixels(bmp2);
            int centerX = bmp2.Width / 2;
            int centerY = bmp2.Height / 2;
            pixels.Sort((p1, p2) =>
            {
                Vector2 p1_v = new Vector2(Math.Abs(p1.X - centerX), Math.Abs(p1.Y - centerY));
                Vector2 p2_v = new Vector2(Math.Abs(p2.X - centerX), Math.Abs(p2.Y - centerY));
                double d1 = p1_v.Diagonal();
                double d2 = p2_v.Diagonal();
                if (d1 > d2)
                    return 1;
                else if (d1 == d2)
                    return 0;
                return -1;
            });
            Console.WriteLine("Calculating...");
            int k = 0;
            Stopwatch sw = Stopwatch.StartNew();
            for (int i = 0; i < pixels.Count; i++)
            {
                if (i % 100 == 0 && i != 0)
                {
                    float percentage = ((float)i / (float)pixels.Count) * 100;
                    Console.WriteLine(percentage.ToString("0.00") + "% completed(" + i + "/" + pixels.Count + ")");
                    Console.SetCursorPosition(0, Console.CursorTop - 1);
                }
                Color set = colors.FindBestColor(pixels[i].Color);
                pixels[i].Color = set;
                k++;
            }
            sw.Stop();
            Console.WriteLine("Saving...");
            Bitmap result = WritePixelsToBitmap(pixels, bmp2.Width, bmp2.Height);
            result.Save(img1 + ".png");
            Console.WriteLine("Completed in " + sw.Elapsed.TotalSeconds + " seconds. Press a key to exit.");
            Console.ReadKey();
        }
        static Bitmap WritePixelsToBitmap(List<Pixel> pixels, int width, int height)
        {
            Bitmap bmp = new Bitmap(width, height);
            foreach (Pixel pixel in pixels)
            {
                bmp.SetPixel(pixel.X, pixel.Y, pixel.Color);
            }
            return bmp;
        }

        static ColorCollection GetColors(Bitmap bmp)
        {
            ColorCollection ret = new ColorCollection();
            for (int x = 0; x < bmp.Width; x++)
            {
                for (int y = 0; y < bmp.Height; y++)
                {
                    Color clr = bmp.GetPixel(x, y);
                    ret.AddColor(clr);
                }
            }
            return ret;
        }
        static List<Pixel> GetPixels(Bitmap bmp)
        {
            List<Pixel> ret = new List<Pixel>();
            for (int x = 0; x < bmp.Width; x++)
            {
                for (int y = 0; y < bmp.Height; y++)
                {
                    Color clr = bmp.GetPixel(x, y);
                    ret.Add(new Pixel(x, y, clr));
                }
            }
            return ret;
        }
    }
}

3

Decidi tentar usar um algoritmo muito semelhante à minha outra resposta, mas apenas trocando blocos de pixels 2x2 em vez de pixels individuais. Infelizmente, esse algoritmo adiciona uma restrição adicional de exigir que as dimensões da imagem sejam divisíveis por 2, o que torna as esferas traçadas por raios inutilizáveis, a menos que eu mude os tamanhos.

Eu realmente gosto de alguns dos resultados!

Gótico americano com paleta rio:

insira a descrição da imagem aqui

Mona Lisa com paleta gótica americana:

insira a descrição da imagem aqui

Mona Lisa com paleta rio:

insira a descrição da imagem aqui

Eu tentei 4x4 também, e aqui estão os meus favoritos!

Noite estrelada com a paleta Scream:

insira a descrição da imagem aqui

Mona Lisa com paleta gótica americana:

insira a descrição da imagem aqui

A paleta Grito com a Mona Lisa:

insira a descrição da imagem aqui

Gótico americano com a paleta Mona Lisa:

insira a descrição da imagem aqui


1
Estava pensando em fazer a mesma coisa + calcular pesos de pixels com base em blocos quadrados. Eu gosto muito dos resultados da Mona Lisa - eles me lembram a imagem da coisa das imagens. Você pode, por acaso, fazer blocos 4x4?
eithed 13/07/2014

1
@eithedog Eu tentei 4x4 e parece muito bom. Veja minha resposta atualizada!
Lvben

3

C #

Isso é realmente muito lento, mas faz um ótimo trabalho, especialmente ao usar a paleta de esferas traçadas por raios.

insira a descrição da imagem aqui insira a descrição da imagem aqui insira a descrição da imagem aqui insira a descrição da imagem aqui insira a descrição da imagem aqui

A paleta Scream:

insira a descrição da imagem aqui insira a descrição da imagem aqui

Paleta Mona Lisa:

insira a descrição da imagem aqui insira a descrição da imagem aqui insira a descrição da imagem aqui insira a descrição da imagem aqui

Paleta gótica americana:

insira a descrição da imagem aqui insira a descrição da imagem aqui

Paleta River:

insira a descrição da imagem aqui insira a descrição da imagem aqui insira a descrição da imagem aqui

A paleta Noite Estrelada:

insira a descrição da imagem aqui insira a descrição da imagem aqui

   class Program
   {
      class Pixel
      {
         public int x;
         public int y;
         public Color color;
         public Pixel(int x, int y, Color color)
         {
            this.x = x;
            this.y = y;
            this.color = color;
         }
      }

      static Pixel BaselineColor = new Pixel(0, 0, Color.FromArgb(0, 0, 0, 0));

      static void Main(string[] args)
      {
         string sourceDirectory = "pic" + args[0] + ".png";
         string paletteDirectory = "pic" + args[1] + ".png";

         using (Bitmap source = Bitmap.FromFile(sourceDirectory) as Bitmap)
         {
            List<Pixel> sourcePixels = GetPixels(source).ToList();
            LinkedList<Pixel> palettePixels;

            using (Bitmap palette = Bitmap.FromFile(paletteDirectory) as Bitmap)
            {
               palettePixels = GetPixels(palette) as LinkedList<Pixel>;
            }

            if (palettePixels.Count != sourcePixels.Count)
            {
               throw new Exception("OH NO!!!!!!!!");
            }

            sourcePixels.Sort((x, y) => GetDiff(y, BaselineColor) - GetDiff(x, BaselineColor));

            LinkedList<Pixel> newPixels = new LinkedList<Pixel>();
            foreach (Pixel p in sourcePixels)
            {
               Pixel newPixel = GetClosestColor(palettePixels, p);
               newPixels.AddLast(newPixel);
            }

            foreach (var p in newPixels)
            {
               source.SetPixel(p.x, p.y, p.color);
            }
            source.Save("Out" + args[0] + "to" + args[1] + ".png");
         }
      }

      private static IEnumerable<Pixel> GetPixels(Bitmap source)
      {
         List<Pixel> newList = new List<Pixel>();
         for (int x = 0; x < source.Width; x++)
         {
            for (int y = 0; y < source.Height; y++)
            {
               newList.Add(new Pixel(x, y, source.GetPixel(x, y)));
            }
         }
         return newList;
      }

      private static Pixel GetClosestColor(LinkedList<Pixel> palettePixels, Pixel p)
      {
         Pixel minPixel = palettePixels.First();
         int diff = GetDiff(minPixel, p);
         foreach (var pix in palettePixels)
         {
            int current = GetDiff(pix, p);
            if (current < diff)
            {
               diff = current;
               minPixel = pix;
               if (diff == 0)
               {
                  return minPixel;
               }
            }
         }
         palettePixels.Remove(minPixel);
         return new Pixel(p.x, p.y, minPixel.color);
      }

      private static int GetDiff(Pixel a, Pixel p)
      {
         return GetProx(a.color, p.color);
      }

      private static int GetProx(Color a, Color p)
      {
         int red = (a.R - p.R) * (a.R - p.R);
         int green = (a.G - p.G) * (a.G - p.G);
         int blue = (a.B - p.B) * (a.B - p.B);
         return red + blue + green;
      }
   }

3

Java - Outra Abordagem de Mapeamento

Edit 1: Depois que isso foi compartilhado em um ambiente de "matemática" no G +, todos nós parecemos usar abordagens correspondentes de várias maneiras para contornar a complexidade.

Edição 2: eu estraguei as imagens no meu google drive e reiniciei, para que os links antigos não funcionem mais. Desculpe, ainda estou trabalhando em mais reputação por mais links.

Edição 3: Lendo as outras postagens, recebi algumas inspirações. Eu peguei o programa mais rápido agora e reinvesti algum tempo de CPU, para fazer algumas alterações, dependendo da localização da imagem de destino.

Edição 4: Nova versão do programa. Mais rápido! Tratamento especial de ambas as áreas com cantos afiados e mudanças muito suaves (ajuda muito no traçado de raios, mas dá ocasionalmente à Mona Lisa olhos vermelhos)! Capacidade de gerar quadros intermediários a partir de animações!

Eu realmente amei a idéia e a solução Quincunx me intrigou. Por isso, pensei que poderia adicionar meus 2 centavos de euro.

A idéia era que, obviamente, precisamos de um mapeamento (de alguma forma próximo) entre duas paletas de cores.

Com essa idéia, passei a primeira noite tentando criar um algoritmo estável de casamento para rodar rápido e com a memória do meu PC em 123520 candidatos. Enquanto eu chegava ao intervalo de memória, achei o problema de tempo de execução insolúvel.

Segunda noite, eu decidi me aprofundar e mergulhar no algoritmo húngaro, que prometia fornecer propriedades de aproximação uniformes, ou seja, distância mínima entre as cores em qualquer imagem. Felizmente, eu encontrei três prontas para implementar implementações Java disso (sem contar muitas tarefas semi-acabadas dos alunos, que começam a dificultar muito o google para algoritmos elementares). Mas, como seria de esperar, os algoritmos húngaros são ainda piores em termos de tempo de execução e uso de memória. Pior ainda, todas as três implementações que testei, retornaram resultados errados ocasionais. Tremo quando penso em outros programas, que podem ser baseados neles.

Terceira abordagem (final da segunda noite) foi fácil, rápida, rápida e, afinal, não é tão ruim assim: classifique as cores nas duas imagens por luminosidade e simples mapa por classificação, ou seja, mais escuro para mais escuro, segundo mais escuro para o segundo mais escuro. Isso cria imediatamente uma reconstrução em preto e branco de aparência nítida, com algumas cores aleatórias espalhadas.

* A abordagem 4 e final até agora (manhã da segunda noite) começa com o mapeamento de luminosidade acima e adiciona correções locais, aplicando algoritmos húngaros a várias seqüências de pixels sobrepostas. Dessa maneira, obtive um mapeamento melhor e trabalhei com a complexidade do problema e os bugs nas implementações.

Então, aqui está um código Java, algumas partes podem parecer semelhantes a outros códigos Java postados aqui. O húngaro usado é uma versão remendada de John Millers, originalmente no projeto ontologySimilariy. Esta foi a maneira mais rápida que encontrei e mostrou o menor número de bugs.

import java.awt.image.BufferedImage;
import java.io.File;
import java.util.ArrayList;
import java.util.Collections;
import java.util.Comparator;
import java.util.Set;
import java.util.HashSet;
import java.util.Map;
import java.util.HashMap;
import java.util.List;
import javax.imageio.ImageIO;

/**
 *
 */
public class PixelRearranger {

    private final String mode;

    public PixelRearranger(String mode)
    {
        this.mode = mode;
    }

    public final static class Pixel {
        final BufferedImage img;
        final int val;
        final int r, g, b;
        final int x, y;

        public Pixel(BufferedImage img, int x, int y) {
            this.x = x;
            this.y = y;
            this.img = img;
            if ( img != null ) {
                val = img.getRGB(x,y);
                r = ((val & 0xFF0000) >> 16);
                g = ((val & 0x00FF00) >> 8);
                b = ((val & 0x0000FF));
            } else {
                val = r = g = b = 0;
            }
        }

        @Override
        public int hashCode() {
            return x + img.getWidth() * y + img.hashCode();
        }

        @Override
        public boolean equals(Object o) {
            if ( !(o instanceof Pixel) ) return false;
            Pixel p2 = (Pixel) o;
            return p2.x == x && p2.y == y && p2.img == img;
        }

        public double cd() {
            double x0 = 0.5 * (img.getWidth()-1);
            double y0 = 0.5 * (img.getHeight()-1);
            return Math.sqrt(Math.sqrt((x-x0)*(x-x0)/x0 + (y-y0)*(y-y0)/y0));
        }

        @Override
        public String toString() { return "P["+r+","+g+","+b+";"+x+":"+y+";"+img.getWidth()+":"+img.getHeight()+"]"; }
    }

    public final static class Pair
        implements Comparable<Pair>
    {   
        public Pixel palette, from;
        public double d;

        public Pair(Pixel palette, Pixel from)
        {
            this.palette = palette;
            this.from = from;
            this.d = distance(palette, from);
        }

        @Override
        public int compareTo(Pair e2)
        {
            return sgn(e2.d - d);
        }

        @Override
        public String toString() { return "E["+palette+from+";"+d+"]"; }
    }

    public static int sgn(double d) { return d > 0.0 ? +1 : d < 0.0 ? -1 : 0; }

    public final static int distance(Pixel p, Pixel q)
    {
        return 3*(p.r-q.r)*(p.r-q.r) + 6*(p.g-q.g)*(p.g-q.g) + (p.b-q.b)*(p.b-q.b);
    }

    public final static Comparator<Pixel> LUMOSITY_COMP = (p1,p2) -> 3*(p1.r-p2.r)+6*(p1.g-p2.g)+(p1.b-p2.b);


    public final static class ArrangementResult
    {
        private List<Pair> pairs;

        public ArrangementResult(List<Pair> pairs)
        {
            this.pairs = pairs;
        }

        /** Provide the output image */
        public BufferedImage finalImage()
        {
            BufferedImage target = pairs.get(0).from.img;
            BufferedImage res = new BufferedImage(target.getWidth(),
                target.getHeight(), BufferedImage.TYPE_INT_RGB);
            for(Pair p : pairs) {
                Pixel left = p.from;
                Pixel right = p.palette;
                res.setRGB(left.x, left.y, right.val);
            }
            return res;
        }

        /** Provide an interpolated image. 0 le;= alpha le;= 1 */
        public BufferedImage interpolateImage(double alpha)
        {
            BufferedImage target = pairs.get(0).from.img;
            int wt = target.getWidth(), ht = target.getHeight();
            BufferedImage palette = pairs.get(0).palette.img;
            int wp = palette.getWidth(), hp = palette.getHeight();
            int w = Math.max(wt, wp), h = Math.max(ht, hp);
            BufferedImage res = new BufferedImage(w, h, BufferedImage.TYPE_INT_RGB);
            int x0t = (w-wt)/2, y0t = (h-ht)/2;
            int x0p = (w-wp)/2, y0p = (h-hp)/2;
            double a0 = (3.0 - 2.0*alpha)*alpha*alpha;
            double a1 = 1.0 - a0;
            for(Pair p : pairs) {
                Pixel left = p.from;
                Pixel right = p.palette;
                int x = (int) (a1 * (right.x + x0p) + a0 * (left.x + x0t));
                int y = (int) (a1 * (right.y + y0p) + a0 * (left.y + y0t));
                if ( x < 0 || x >= w ) System.out.println("x="+x+", w="+w+", alpha="+alpha);
                if ( y < 0 || y >= h ) System.out.println("y="+y+", h="+h+", alpha="+alpha);
                res.setRGB(x, y, right.val);
            }
            return res;
        }
    }

    public ArrangementResult rearrange(BufferedImage target, BufferedImage palette)
    {
        List<Pixel> targetPixels = getColors(target);
        int n = targetPixels.size();
        System.out.println("total Pixels "+n);
        Collections.sort(targetPixels, LUMOSITY_COMP);

        final double[][] energy = energy(target);

        List<Pixel> palettePixels = getColors(palette);
        Collections.sort(palettePixels, LUMOSITY_COMP);

        ArrayList<Pair> pairs = new ArrayList<>(n);
        for(int i = 0; i < n; i++) {
            Pixel pal = palettePixels.get(i);
            Pixel to = targetPixels.get(i);
            pairs.add(new Pair(pal, to));
        }
        correct(pairs, (p1,p2) -> sgn(p2.d*p2.from.b - p1.d*p1.from.b));
        correct(pairs, (p1,p2) -> sgn(p2.d*p2.from.r - p1.d*p1.from.r));
        // generates visible circular artifacts: correct(pairs, (p1,p2) -> sgn(p2.d*p2.from.cd() - p1.d*p1.from.cd()));
        correct(pairs, (p1,p2) -> sgn(energy[p2.from.x][p2.from.y]*p2.d - energy[p1.from.x][p1.from.y]*p1.d));
        correct(pairs, (p1,p2) -> sgn(p2.d/(1+energy[p2.from.x][p2.from.y]) - p1.d/(1+energy[p1.from.x][p1.from.y])));
        // correct(pairs, null);
        return new ArrangementResult(pairs);

    }

    /**
     * derive an energy map, to detect areas of lots of change.
     */
    public double[][] energy(BufferedImage img)
    {
        int n = img.getWidth();
        int m = img.getHeight();
        double[][] res = new double[n][m];
        for(int x = 0; x < n; x++) {
            for(int y = 0; y < m; y++) {
                int rgb0 = img.getRGB(x,y);
                int count = 0, sum = 0;
                if ( x > 0 ) {
                    count++; sum += dist(rgb0, img.getRGB(x-1,y));
                    if ( y > 0 ) { count++; sum += dist(rgb0, img.getRGB(x-1,y-1)); }
                    if ( y < m-1 ) { count++; sum += dist(rgb0, img.getRGB(x-1,y+1)); }
                }
                if ( x < n-1 ) {
                    count++; sum += dist(rgb0, img.getRGB(x+1,y));
                    if ( y > 0 ) { count++; sum += dist(rgb0, img.getRGB(x+1,y-1)); }
                    if ( y < m-1 ) { count++; sum += dist(rgb0, img.getRGB(x+1,y+1)); }
                }
                if ( y > 0 ) { count++; sum += dist(rgb0, img.getRGB(x,y-1)); }
                if ( y < m-1 ) { count++; sum += dist(rgb0, img.getRGB(x,y+1)); }
                res[x][y] = Math.sqrt((double)sum/count);
            }
        }
        return res;
    }

    public int dist(int rgb0, int rgb1) {
        int r0 = ((rgb0 & 0xFF0000) >> 16);
        int g0 = ((rgb0 & 0x00FF00) >> 8);
        int b0 = ((rgb0 & 0x0000FF));
        int r1 = ((rgb1 & 0xFF0000) >> 16);
        int g1 = ((rgb1 & 0x00FF00) >> 8);
        int b1 = ((rgb1 & 0x0000FF));
        return 3*(r0-r1)*(r0-r1) + 6*(g0-g1)*(g0-g1) + (b0-b1)*(b0-b1);
    }

    private void correct(ArrayList<Pair> pairs, Comparator<Pair> comp)
    {
        Collections.sort(pairs, comp);
        int n = pairs.size();
        int limit = Math.min(n, 133); // n / 1000;
        int limit2 = Math.max(1, n / 3 - limit);
        int step = (2*limit + 2)/3;
        for(int base = 0; base < limit2; base += step ) {
            List<Pixel> list1 = new ArrayList<>();
            List<Pixel> list2 = new ArrayList<>();
            for(int i = base; i < base+limit; i++) {
                list1.add(pairs.get(i).from);
                list2.add(pairs.get(i).palette);
            }
            Map<Pixel, Pixel> connection = rematch(list1, list2);
            int i = base;
            for(Pixel p : connection.keySet()) {
                pairs.set(i++, new Pair(p, connection.get(p)));
            }
        }
    }

    /**
     * Glue code to do an hungarian algorithm distance optimization.
     */
    public Map<Pixel,Pixel> rematch(List<Pixel> liste1, List<Pixel> liste2)
    {
        int n = liste1.size();
        double[][] cost = new double[n][n];
        Set<Pixel> s1 = new HashSet<>(n);
        Set<Pixel> s2 = new HashSet<>(n);
        for(int i = 0; i < n; i++) {
            Pixel ii = liste1.get(i);
            for(int j = 0; j < n; j++) {
                Pixel ij = liste2.get(j);
                cost[i][j] = -distance(ii,ij);
            }
        }
        Map<Pixel,Pixel> res = new HashMap<>();
        int[] resArray = Hungarian.hungarian(cost);
        for(int i = 0; i < resArray.length; i++) {
            Pixel ii = liste1.get(i);
            Pixel ij = liste2.get(resArray[i]);
            res.put(ij, ii);
        }
        return res;
    }

    public static List<Pixel> getColors(BufferedImage img) {
        int width = img.getWidth();
        int height = img.getHeight();
        List<Pixel> colors = new ArrayList<>(width * height);
        for (int x = 0; x < width; x++) {
            for (int y = 0; y < height; y++) {
                colors.add(new Pixel(img, x, y));
            }
        }
        return colors;
    }

    public static List<Integer> getSortedTrueColors(BufferedImage img) {
        int width = img.getWidth();
        int height = img.getHeight();
        List<Integer> colors = new ArrayList<>(width * height);
        for (int x = 0; x < width; x++) {
            for (int y = 0; y < height; y++) {
                colors.add(img.getRGB(x, y));
            }
        }
        Collections.sort(colors);
        return colors;
    }

    public static void main(String[] args) throws Exception {
        int i = 0;
        String mode = args[i++];
        PixelRearranger pr = new PixelRearranger(mode);
        String a1 = args[i++];
        File in1 = new File(a1);
        String a2 = args[i++];
        File in2 = new File(a2);
        File out = new File(args[i++]);
        //
        BufferedImage target = ImageIO.read(in1);
        BufferedImage palette = ImageIO.read(in2);
        long t0 = System.currentTimeMillis();
        ArrangementResult result = pr.rearrange(target, palette);
        BufferedImage resultImg = result.finalImage();
        long t1 = System.currentTimeMillis();
        System.out.println("took "+0.001*(t1-t0)+" s");
        ImageIO.write(resultImg, "png", out);
        // Check validity
        List<Integer> paletteColors = getSortedTrueColors(palette);
        List<Integer> resultColors = getSortedTrueColors(resultImg);
        System.out.println("validate="+paletteColors.equals(resultColors));
        // In Mode A we do some animation!
        if ( "A".equals(mode) ) {
            for(int j = 0; j <= 50; j++) {
                BufferedImage stepImg = result.interpolateImage(0.02 * j);
                File oa = new File(String.format("anim/%s-%s-%02d.png", a1, a2, j));
                ImageIO.write(stepImg, "png", oa);
            }
        }
    }
}

O tempo de execução atual é de 20 a 30 segundos por par de imagens acima, mas há muitos ajustes para torná-lo mais rápido ou talvez obter um pouco mais de qualidade.

Parece que minha reputação de novato não é suficiente para tantos links / imagens, então aqui está um atalho de texto para minha pasta do Google drives para obter exemplos de imagens: http://goo.gl/qZHTao

As amostras que eu queria mostrar primeiro:

Pessoas -> Mona Lisa http://goo.gl/mGvq9h

O programa acompanha todas as coordenadas dos pontos, mas agora me sinto exausto e não pretendo fazer animações por enquanto. Se eu gastasse mais tempo, eu mesmo poderia executar um algoritmo húngaro ou ajustar a programação de otimização local do meu programa.

Ao utilizar nosso site, você reconhece que leu e compreendeu nossa Política de Cookies e nossa Política de Privacidade.
Licensed under cc by-sa 3.0 with attribution required.