Java: intente sin pérdidas y recurra a contenido
(¡El mejor resultado sin pérdidas hasta ahora!)
Cuando examiné por primera vez esta pregunta, pensé que no era un rompecabezas o un desafío, solo alguien que necesitaba desesperadamente un programa y su código;) Pero está en mi naturaleza resolver problemas de visión, por lo que no pude evitar que este intentara este desafío. !
Se me ocurrió el siguiente enfoque y combinación de algoritmos.
En pseudocódigo se ve así:
function crop(image, desired) {
int sizeChange = 1;
while(sizeChange != 0 and image.width > desired){
Look for a repeating and connected set of lines (top to bottom) with a minimum of x lines
Remove all the lines except for one
sizeChange = image.width - newImage.width
image = newImage;
}
if(image.width > desired){
while(image.width > 2 and image.width > desired){
Create a "pixel energy" map of the image
Find the path from the top of the image to the bottom which "costs" the least amount of "energy"
Remove the lowest cost path from the image
image = newImage;
}
}
}
int desiredWidth = ?
int desiredHeight = ?
Image image = input;
crop(image, desiredWidth);
rotate(image, 90);
crop(image, desiredWidth);
rotate(image, -90);
Técnicas utilizadas
- Intensidad en escala de grises
- Dilatación
- Igual columna de búsqueda y eliminar
- Tallado de costuras
- Detección de bordes sobel
- Umbral
El programa
El programa puede recortar capturas de pantalla sin pérdidas, pero tiene una opción para recurrir a recortes conscientes del contenido que no son 100% sin pérdidas. Los argumentos del programa se pueden ajustar para lograr mejores resultados.
Nota: El programa se puede mejorar de muchas maneras (¡no tengo tanto tiempo libre!)
Argumentos
File name = file
Desired width = number > 0
Desired height = number > 0
Min slice width = number > 1
Compare threshold = number > 0
Use content aware = boolean
Max content aware cycles = number >= 0
Código
import java.awt.Graphics2D;
import java.awt.image.BufferedImage;
import java.awt.image.ColorModel;
import java.io.File;
import java.io.IOException;
import javax.imageio.ImageIO;
import javax.swing.ImageIcon;
import javax.swing.JLabel;
import javax.swing.JOptionPane;
/**
* @author Rolf Smit
* Share and adapt as you like, but don't forget to credit the author!
*/
public class MagicWindowCropper {
public static void main(String[] args) {
if(args.length != 7){
throw new IllegalArgumentException("At least 7 arguments are required: (file, desiredWidth, desiredHeight, minSliceSize, sliceThreshold, forceRemove, maxForceRemove)!");
}
File file = new File(args[0]);
int minSliceSize = Integer.parseInt(args[3]); //4;
int desiredWidth = Integer.parseInt(args[1]); //400;
int desiredHeight = Integer.parseInt(args[2]); //400;
boolean forceRemove = Boolean.parseBoolean(args[5]); //true
int maxForceRemove = Integer.parseInt(args[6]); //40
MagicWindowCropper.MATCH_THRESHOLD = Integer.parseInt(args[4]); //3;
try {
BufferedImage result = ImageIO.read(file);
System.out.println("Horizontal cropping");
//Horizontal crop
result = doDuplicateColumnsMagic(result, minSliceSize, desiredWidth);
if (result.getWidth() != desiredWidth && forceRemove) {
result = doSeamCarvingMagic(result, maxForceRemove, desiredWidth);
}
result = getRotatedBufferedImage(result, false);
System.out.println("Vertical cropping");
//Vertical crop
result = doDuplicateColumnsMagic(result, minSliceSize, desiredHeight);
if (result.getWidth() != desiredHeight && forceRemove) {
result = doSeamCarvingMagic(result, maxForceRemove, desiredHeight);
}
result = getRotatedBufferedImage(result, true);
showBufferedImage("Result", result);
ImageIO.write(result, "png", getNewFileName(file));
} catch (IOException e) {
e.printStackTrace();
}
}
private static BufferedImage doSeamCarvingMagic(BufferedImage inputImage, int max, int desired) {
System.out.println("Seam Carving magic:");
int maxChange = Math.min(inputImage.getWidth() - desired, max);
BufferedImage last = inputImage;
int total = 0, change;
do {
int[][] energy = getPixelEnergyImage(last);
BufferedImage out = removeLowestSeam(energy, last);
change = last.getWidth() - out.getWidth();
total += change;
System.out.println("Carves removed: " + total);
last = out;
} while (change != 0 && total < maxChange);
return last;
}
private static BufferedImage doDuplicateColumnsMagic(BufferedImage inputImage, int minSliceWidth, int desired) {
System.out.println("Duplicate columns magic:");
int maxChange = inputImage.getWidth() - desired;
BufferedImage last = inputImage;
int total = 0, change;
do {
BufferedImage out = removeDuplicateColumn(last, minSliceWidth, desired);
change = last.getWidth() - out.getWidth();
total += change;
System.out.println("Columns removed: " + total);
last = out;
} while (change != 0 && total < maxChange);
return last;
}
/*
* Duplicate column methods
*/
private static BufferedImage removeDuplicateColumn(BufferedImage inputImage, int minSliceWidth, int desiredWidth) {
if (inputImage.getWidth() <= minSliceWidth) {
throw new IllegalStateException("The image width is smaller than the minSliceWidth! What on earth are you trying to do?!");
}
int[] stamp = null;
int sliceStart = -1, sliceEnd = -1;
for (int x = 0; x < inputImage.getWidth() - minSliceWidth + 1; x++) {
stamp = getHorizontalSliceStamp(inputImage, x, minSliceWidth);
if (stamp != null) {
sliceStart = x;
sliceEnd = x + minSliceWidth - 1;
break;
}
}
if (stamp == null) {
return inputImage;
}
BufferedImage out = deepCopyImage(inputImage);
for (int x = sliceEnd + 1; x < inputImage.getWidth(); x++) {
int[] row = getHorizontalSliceStamp(inputImage, x, 1);
if (equalsRows(stamp, row)) {
sliceEnd = x;
} else {
break;
}
}
//Remove policy
int canRemove = sliceEnd - (sliceStart + 1) + 1;
int mayRemove = inputImage.getWidth() - desiredWidth;
int dif = mayRemove - canRemove;
if (dif < 0) {
sliceEnd += dif;
}
int mustRemove = sliceEnd - (sliceStart + 1) + 1;
if (mustRemove <= 0) {
return out;
}
out = removeHorizontalRegion(out, sliceStart + 1, sliceEnd);
out = removeLeft(out, out.getWidth() - mustRemove);
return out;
}
private static BufferedImage removeHorizontalRegion(BufferedImage image, int startX, int endX) {
int width = endX - startX + 1;
if (endX + 1 > image.getWidth()) {
endX = image.getWidth() - 1;
}
if (endX < startX) {
throw new IllegalStateException("Invalid removal parameters! Wow this error message is genius!");
}
BufferedImage out = deepCopyImage(image);
for (int x = endX + 1; x < image.getWidth(); x++) {
for (int y = 0; y < image.getHeight(); y++) {
out.setRGB(x - width, y, image.getRGB(x, y));
out.setRGB(x, y, 0xFF000000);
}
}
return out;
}
private static int[] getHorizontalSliceStamp(BufferedImage inputImage, int startX, int sliceWidth) {
int[] initial = new int[inputImage.getHeight()];
for (int y = 0; y < inputImage.getHeight(); y++) {
initial[y] = inputImage.getRGB(startX, y);
}
if (sliceWidth == 1) {
return initial;
}
for (int s = 1; s < sliceWidth; s++) {
int[] row = new int[inputImage.getHeight()];
for (int y = 0; y < inputImage.getHeight(); y++) {
row[y] = inputImage.getRGB(startX + s, y);
}
if (!equalsRows(initial, row)) {
return null;
}
}
return initial;
}
private static int MATCH_THRESHOLD = 3;
private static boolean equalsRows(int[] left, int[] right) {
for (int i = 0; i < left.length; i++) {
int rl = (left[i]) & 0xFF;
int gl = (left[i] >> 8) & 0xFF;
int bl = (left[i] >> 16) & 0xFF;
int rr = (right[i]) & 0xFF;
int gr = (right[i] >> 8) & 0xFF;
int br = (right[i] >> 16) & 0xFF;
if (Math.abs(rl - rr) > MATCH_THRESHOLD
|| Math.abs(gl - gr) > MATCH_THRESHOLD
|| Math.abs(bl - br) > MATCH_THRESHOLD) {
return false;
}
}
return true;
}
/*
* Seam carving methods
*/
private static BufferedImage removeLowestSeam(int[][] input, BufferedImage image) {
int lowestValue = Integer.MAX_VALUE; //Integer overflow possible when image height grows!
int lowestValueX = -1;
// Here be dragons
for (int x = 1; x < input.length - 1; x++) {
int seamX = x;
int value = input[x][0];
for (int y = 1; y < input[x].length; y++) {
if (seamX < 1) {
int top = input[seamX][y];
int right = input[seamX + 1][y];
if (top <= right) {
value += top;
} else {
seamX++;
value += right;
}
} else if (seamX > input.length - 2) {
int top = input[seamX][y];
int left = input[seamX - 1][y];
if (top <= left) {
value += top;
} else {
seamX--;
value += left;
}
} else {
int left = input[seamX - 1][y];
int top = input[seamX][y];
int right = input[seamX + 1][y];
if (top <= left && top <= right) {
value += top;
} else if (left <= top && left <= right) {
seamX--;
value += left;
} else {
seamX++;
value += right;
}
}
}
if (value < lowestValue) {
lowestValue = value;
lowestValueX = x;
}
}
BufferedImage out = deepCopyImage(image);
int seamX = lowestValueX;
shiftRow(out, seamX, 0);
for (int y = 1; y < input[seamX].length; y++) {
if (seamX < 1) {
int top = input[seamX][y];
int right = input[seamX + 1][y];
if (top <= right) {
shiftRow(out, seamX, y);
} else {
seamX++;
shiftRow(out, seamX, y);
}
} else if (seamX > input.length - 2) {
int top = input[seamX][y];
int left = input[seamX - 1][y];
if (top <= left) {
shiftRow(out, seamX, y);
} else {
seamX--;
shiftRow(out, seamX, y);
}
} else {
int left = input[seamX - 1][y];
int top = input[seamX][y];
int right = input[seamX + 1][y];
if (top <= left && top <= right) {
shiftRow(out, seamX, y);
} else if (left <= top && left <= right) {
seamX--;
shiftRow(out, seamX, y);
} else {
seamX++;
shiftRow(out, seamX, y);
}
}
}
return removeLeft(out, out.getWidth() - 1);
}
private static void shiftRow(BufferedImage image, int startX, int y) {
for (int x = startX; x < image.getWidth() - 1; x++) {
image.setRGB(x, y, image.getRGB(x + 1, y));
}
}
private static int[][] getPixelEnergyImage(BufferedImage image) {
// Convert Image to gray scale using the luminosity method and add extra
// edges for the Sobel filter
int[][] grayScale = new int[image.getWidth() + 2][image.getHeight() + 2];
for (int x = 0; x < image.getWidth(); x++) {
for (int y = 0; y < image.getHeight(); y++) {
int rgb = image.getRGB(x, y);
int r = (rgb >> 16) & 0xFF;
int g = (rgb >> 8) & 0xFF;
int b = (rgb & 0xFF);
int luminosity = (int) (0.21 * r + 0.72 * g + 0.07 * b);
grayScale[x + 1][y + 1] = luminosity;
}
}
// Sobel edge detection
final double[] kernelHorizontalEdges = new double[] { 1, 2, 1, 0, 0, 0, -1, -2, -1 };
final double[] kernelVerticalEdges = new double[] { 1, 0, -1, 2, 0, -2, 1, 0, -1 };
int[][] energyImage = new int[image.getWidth()][image.getHeight()];
for (int x = 1; x < image.getWidth() + 1; x++) {
for (int y = 1; y < image.getHeight() + 1; y++) {
int k = 0;
double horizontal = 0;
for (int ky = -1; ky < 2; ky++) {
for (int kx = -1; kx < 2; kx++) {
horizontal += ((double) grayScale[x + kx][y + ky] * kernelHorizontalEdges[k]);
k++;
}
}
double vertical = 0;
k = 0;
for (int ky = -1; ky < 2; ky++) {
for (int kx = -1; kx < 2; kx++) {
vertical += ((double) grayScale[x + kx][y + ky] * kernelVerticalEdges[k]);
k++;
}
}
if (Math.sqrt(horizontal * horizontal + vertical * vertical) > 127) {
energyImage[x - 1][y - 1] = 255;
} else {
energyImage[x - 1][y - 1] = 0;
}
}
}
//Dilate the edge detected image a few times for better seaming results
//Current value is just 1...
for (int i = 0; i < 1; i++) {
dilateImage(energyImage);
}
return energyImage;
}
private static void dilateImage(int[][] image) {
for (int x = 0; x < image.length; x++) {
for (int y = 0; y < image[x].length; y++) {
if (image[x][y] == 255) {
if (x > 0 && image[x - 1][y] == 0) {
image[x - 1][y] = 2; //Note: 2 is just a placeholder value
}
if (y > 0 && image[x][y - 1] == 0) {
image[x][y - 1] = 2;
}
if (x + 1 < image.length && image[x + 1][y] == 0) {
image[x + 1][y] = 2;
}
if (y + 1 < image[x].length && image[x][y + 1] == 0) {
image[x][y + 1] = 2;
}
}
}
}
for (int x = 0; x < image.length; x++) {
for (int y = 0; y < image[x].length; y++) {
if (image[x][y] == 2) {
image[x][y] = 255;
}
}
}
}
/*
* Utilities
*/
private static void showBufferedImage(String windowTitle, BufferedImage image) {
JOptionPane.showMessageDialog(null, new JLabel(new ImageIcon(image)), windowTitle, JOptionPane.PLAIN_MESSAGE, null);
}
private static BufferedImage deepCopyImage(BufferedImage input) {
ColorModel cm = input.getColorModel();
return new BufferedImage(cm, input.copyData(null), cm.isAlphaPremultiplied(), null);
}
private static final BufferedImage getRotatedBufferedImage(BufferedImage img, boolean back) {
double oldW = img.getWidth(), oldH = img.getHeight();
double newW = img.getHeight(), newH = img.getWidth();
BufferedImage out = new BufferedImage((int) newW, (int) newH, img.getType());
Graphics2D g = out.createGraphics();
g.translate((newW - oldW) / 2.0, (newH - oldH) / 2.0);
g.rotate(Math.toRadians(back ? -90 : 90), oldW / 2.0, oldH / 2.0);
g.drawRenderedImage(img, null);
g.dispose();
return out;
}
private static BufferedImage removeLeft(BufferedImage image, int startX) {
int removeWidth = image.getWidth() - startX;
BufferedImage out = new BufferedImage(image.getWidth() - removeWidth,
image.getHeight(), image.getType());
for (int x = 0; x < startX; x++) {
for (int y = 0; y < out.getHeight(); y++) {
out.setRGB(x, y, image.getRGB(x, y));
}
}
return out;
}
private static File getNewFileName(File in) {
String name = in.getName();
int i = name.lastIndexOf(".");
if (i != -1) {
String ext = name.substring(i);
String n = name.substring(0, i);
return new File(in.getParentFile(), n + "-cropped" + ext);
} else {
return new File(in.getParentFile(), name + "-cropped");
}
}
}
Resultados
Captura de pantalla XP sin pérdida sin el tamaño deseado (compresión máxima sin pérdida)
Argumentos: "image.png" 1 1 5 10 falso 0
Resultado: 836 x 323
Captura de pantalla de XP a 800x600
Argumentos: "image.png" 800600 6 10 verdadero 60
Resultado: 800 x 600
El algoritmo sin pérdida elimina alrededor de 155 líneas horizontales de las que el algoritmo recurre a la eliminación consciente del contenido para que se puedan ver algunos artefactos.
Captura de pantalla de Windows 10 a 700x300
Argumentos: "image.png" 700300 6 10 verdadero 60
Resultado: 700 x 300
El algoritmo sin pérdida elimina 270 líneas horizontales de las que el algoritmo recurre a la eliminación de contenido que elimina otras 29. Vertical solo se utiliza el algoritmo sin pérdida.
Captura de pantalla de Windows 10 con reconocimiento de contenido a 400x200 (prueba)
Argumentos: "image.png" 400200 5 10 verdadero 600
Resultado: 400 x 200
Esta fue una prueba para ver cómo se vería la imagen resultante después del uso severo de la función de contenido. El resultado está muy dañado pero no es irreconocible.
