/*
* @(#)WhiteBalance.java
*
* Copyright (c) 2012 Werner Randelshofer, Goldau, Switzerland.
* All rights reserved.
*
* You may not use, copy or modify this file, except in compliance with the
* license agreement you entered into with Werner Randelshofer.
* For details see accompanying license terms.
*/
package org.monte.media.image;
import java.util.Arrays;
import org.monte.media.math.LinearEquations;
import java.awt.image.BufferedImage;
import java.awt.image.DataBufferInt;
import javax.media.jai.Histogram;
import static java.lang.Math.*;
/**
* {@code WhiteBalance}.
* <p>
* References:<br>
* [Ken09] Kenfack, Pierre Marie (2009). Implementierung und Vergleich
* verschiedener Algorithmen zur Bildsensorkalibrierung. Fraunhofer ITWM.
* http://www.itwm.fraunhofer.de/fileadmin/ITWM-Media/Abteilungen/BV/Pdf/
Diplomarbeit_Kenfack.pdf
* <p>
* [Lam05] Edmund Lam, Combining gray world and retinex theory for automatic
* white balance in digital photography, Consumer Electronics, 2005.
* (ISCE 2005). Proceedings of the Ninth International Symposium on (2005), pp.134–139.
* <p>
* [Huo05] Huo Yun-yan, Chang Yi-lin, Wang Jing, Wei Xiao-xia. (2005). Robust
* Automatic White Balance Algorithm using Gray Color Points in Images.
*
* @author Werner Randelshofer
* @version $Id: WhiteBalance.java 299 2013-01-03 07:40:18Z werner $
*/
public class WhiteBalance {
private WhiteBalance() {
}
;
/** Performs white balance adjustment using the "grey world" assumption
* as described in [Ken09]. */
public static BufferedImage whiteBalanceGreyworld(BufferedImage img) {
img = Images.toIntImage(img);
Histogram hist = new Histogram(256, 0, 255, 3);
hist.countPixels(img.getRaster(), null, 0, 0, 1, 1);
double[] m = whiteBalanceGreyworld(hist);
BufferedImage out = new BufferedImage(img.getWidth(), img.getHeight(), BufferedImage.TYPE_INT_RGB);
int[] p = ((DataBufferInt) img.getRaster().getDataBuffer()).getData();
int[] q = ((DataBufferInt) out.getRaster().getDataBuffer()).getData();
for (int i = 0; i < p.length; i++) {
int px = p[i];
double R = (px & 0xff0000) >> 16;
double G = (px & 0xff00) >> 8;
double B = (px & 0xff) >> 0;
double Rq = m[0] * R + m[1] * B + m[2] * G;
double Gq = m[3] * R + m[4] * B + m[5] * G;
double Bq = m[6] * R + m[7] * B + m[8] * G;
q[i] = ((min(255, max(0, (int) Rq))) & 0xff) << 16
| ((min(255, max(0, (int) Gq))) & 0xff) << 8
| ((min(255, max(0, (int) Bq))) & 0xff) << 0;
}
return out;
}
/** Performs white balance adjustment using the "grey world" assumption
* as described in [Huo05], but using the YCbCr color space instead of YUV. */
public static BufferedImage whiteBalanceGreyworldYCC(BufferedImage img, float[] ccAdjust, boolean all) {
img = Images.toIntImage(img);
Histogram hist = new Histogram(256, 0, 255, 3);
hist.countPixels(img.getRaster(), null, 0, 0, 1, 1);
double[] m = whiteBalanceGreyworld(hist);
int[] p = ((DataBufferInt) img.getRaster().getDataBuffer()).getData();
float[] rgb = new float[3];
float[] ycc = new float[3];
float T = 0.097f;
int NGray = 0, NColor = 0;
double cbGraySum = 0, cbColorSum = 0;
double crGraySum = 0, crColorSum = 0;
for (int i = 0; i < p.length; i++) {
int px = p[i];
rgb[0] = ((px & 0xff0000) >> 16) / 255f;
rgb[1] = ((px & 0xff00) >> 8) / 255f;
rgb[2] = ((px & 0xff) >> 0) / 255f;
ColorModels.RGBtoYCC(rgb, ycc);
if ((abs(ycc[1]) + abs(ycc[2])) / ycc[0] > T) {
NColor++;
cbColorSum += ycc[1];
crColorSum += ycc[2];
} else {
NGray++;
cbGraySum += ycc[1];
crGraySum += ycc[2];
}
}
float cbGrayAdj = -(float) (cbGraySum / NGray);
float crGrayAdj = -(float) (crGraySum / NGray);
float cbAllAdj = -(float) ((cbColorSum / NColor) + cbGraySum / NGray);
float crAllAdj = -(float) (crColorSum / NColor + crGraySum / NGray);
float cbAdj = cbGrayAdj;
float crAdj = crGrayAdj;
if (all) {
cbAdj = cbAllAdj;
crAdj = crAllAdj;
}
System.out.println("WhiteBalance.YCC GRAY cb=" + cbGrayAdj + " cr=" + crGrayAdj + " N=" + NGray);
System.out.println("WhiteBalance.YCC ALL cb=" + cbAllAdj + " cr=" + crAllAdj + " N=" + NColor);
if (ccAdjust != null) {
ccAdjust[0] = cbAdj;
ccAdjust[1] = crAdj;
return null;
} else {
BufferedImage out = new BufferedImage(img.getWidth(), img.getHeight(), BufferedImage.TYPE_INT_RGB);
int[] q = ((DataBufferInt) out.getRaster().getDataBuffer()).getData();
for (int i = 0; i < p.length; i++) {
int px = p[i];
rgb[0] = ((px & 0xff0000) >> 16) / 255f;
rgb[1] = ((px & 0xff00) >> 8) / 255f;
rgb[2] = ((px & 0xff) >> 0) / 255f;
ColorModels.RGBtoYCC(rgb, ycc);
ycc[1] += cbAdj;
ycc[2] += crAdj;
ColorModels.YCCtoRGB(ycc, rgb);
q[i] = ((int) (rgb[0] * 255) << 16)//
| ((int) (rgb[1] * 255) << 8)//
| ((int) (rgb[2] * 255) << 0);
}
return out;
}
}
/** Performs white balance adjustment using the "grey world" assumption
* as described in [Huo05]. */
public static BufferedImage whiteBalanceGreyworldYUV(BufferedImage img, float[] uvAdjust, boolean all) {
img = Images.toIntImage(img);
Histogram hist = new Histogram(256, 0, 255, 3);
hist.countPixels(img.getRaster(), null, 0, 0, 1, 1);
double[] m = whiteBalanceGreyworld(hist);
int[] p = ((DataBufferInt) img.getRaster().getDataBuffer()).getData();
float[] rgb = new float[3];
float[] yuv = new float[3];
float T = 0.097f;
int NGray = 0, NColor = 0;
double UGraySum = 0, UColorSum = 0;
double VGraySum = 0, VColorSum = 0;
for (int i = 0; i < p.length; i++) {
int px = p[i];
rgb[0] = ((px & 0xff0000) >> 16) / 255f;
rgb[1] = ((px & 0xff00) >> 8) / 255f;
rgb[2] = ((px & 0xff) >> 0) / 255f;
ColorModels.RGBtoYUV(rgb, yuv);
if ((abs(yuv[1]) + abs(yuv[2])) / yuv[0] > T) {
NColor++;
UColorSum += yuv[1];
VColorSum += yuv[2];
} else {
NGray++;
UGraySum += yuv[1];
VGraySum += yuv[2];
}
}
float UGrayAdj = -(float) (UGraySum / NGray);
float VGrayAdj = -(float) (VGraySum / NGray);
float UAllAdj = -(float) ((UColorSum / NColor) + UGraySum / NGray);
float VAllAdj = -(float) (VColorSum / NColor + VGraySum / NGray);
float UAdj = UGrayAdj;
float VAdj = VGrayAdj;
if (all) {
UAdj = UAllAdj;
VAdj = VAllAdj;
}
System.out.println("WhiteBalance.YUV GRAY cb=" + UGrayAdj + " cr=" + VGrayAdj + " N=" + NGray);
System.out.println("WhiteBalance.YUV ALL cb=" + UAllAdj + " cr=" + VAllAdj + " N=" + NColor);
if (uvAdjust != null) {
uvAdjust[0] = UAdj;
uvAdjust[1] = VAdj;
return null;
} else {
BufferedImage out = new BufferedImage(img.getWidth(), img.getHeight(), BufferedImage.TYPE_INT_RGB);
int[] q = ((DataBufferInt) out.getRaster().getDataBuffer()).getData();
for (int i = 0; i < p.length; i++) {
int px = p[i];
rgb[0] = ((px & 0xff0000) >> 16) / 255f;
rgb[1] = ((px & 0xff00) >> 8) / 255f;
rgb[2] = ((px & 0xff) >> 0) / 255f;
ColorModels.RGBtoYUV(rgb, yuv);
yuv[1] += UAdj;
yuv[2] += VAdj;
ColorModels.YUVtoRGB(yuv, rgb);
q[i] = ((int) (rgb[0] * 255) << 16)//
| ((int) (rgb[1] * 255) << 8)//
| ((int) (rgb[2] * 255) << 0);
}
return out;
}
}
public static BufferedImage whiteBalanceRetinex(BufferedImage img) {
img = Images.toIntImage(img);
Histogram hist = new Histogram(256, 0, 255, 3);
hist.countPixels(img.getRaster(), null, 0, 0, 1, 1);
double[] m = whiteBalanceRetinex(hist);
BufferedImage out = new BufferedImage(img.getWidth(), img.getHeight(), BufferedImage.TYPE_INT_RGB);
int[] p = ((DataBufferInt) img.getRaster().getDataBuffer()).getData();
int[] q = ((DataBufferInt) out.getRaster().getDataBuffer()).getData();
for (int i = 0; i < p.length; i++) {
int px = p[i];
double R = (px & 0xff0000) >> 16;
double G = (px & 0xff00) >> 8;
double B = (px & 0xff) >> 0;
double Rq = m[0] * R + m[1] * B + m[2] * G;
double Gq = m[3] * R + m[4] * B + m[5] * G;
double Bq = m[6] * R + m[7] * B + m[8] * G;
q[i] = ((max(0, (int) Rq)) & 0xff) << 16
| ((max(0, (int) Gq)) & 0xff) << 8
| ((max(0, (int) Bq)) & 0xff) << 0;
}
return out;
}
public static BufferedImage whiteBalanceQM(BufferedImage img) {
img = Images.toIntImage(img);
Histogram hist = new Histogram(256, 0, 255, 3);
hist.countPixels(img.getRaster(), null, 0, 0, 1, 1);
double[] m = whiteBalanceQM(hist);
BufferedImage out = new BufferedImage(img.getWidth(), img.getHeight(), BufferedImage.TYPE_INT_RGB);
int[] p = ((DataBufferInt) img.getRaster().getDataBuffer()).getData();
int[] q = ((DataBufferInt) out.getRaster().getDataBuffer()).getData();
float mur = (float) m[0];
float nur = (float) m[1];
float mub = (float) m[2];
float nub = (float) m[3];
System.out.println("WhiteBalance QM mur=" + mur + " nur=" + nur+" mub=" + mub + " nub=" + nub);
for (int i = 0; i < p.length; i++) {
int px = p[i];
double R = (px & 0xff0000) >> 16;
double B = (px & 0xff) >> 0;
double Rq = mur * R * R + nur * R;
double Bq = mub * B * B + nub * B;
q[i] = ((min(255, max(0, (int) Rq)))) << 16
| (px & 0xff00)
| ((min(255, max(0, (int) Bq)))) << 0;
}
return out;
}
/**
* Computes the white balance of an image using the Greyworld algorithm.
* <p>
* The Greyworld algorithm assumes that the average color of an image
* should be a neutral gray: avg(R)=avg(G)=avg(B).
* <p>
* References:<br>
* Edmund Lam, Combining gray world and retinex theory for automatic white
* balance in digital photography, Consumer Electronics, 2005. (ISCE 2005).
* Proceedings of the Ninth International Symposium on (2005), pp.134–139.
* <p>
* Kenfack, Pierre Marie. (2009). Implementierung und Vergleich verschiedener
* Algorithmen zur Bildsensorkalibrierung. Fraunhofer ITWM.
* http://www.itwm.fraunhofer.de/fileadmin/ITWM
* <p>
*
*
* @param rgbHist
* @return a 3x3 matrix which performs the color correction matrix*[R,G,B].
*/
public static double[] whiteBalanceGreyworld(Histogram rgbHist) {
double[] mean_ = rgbHist.getMean();
double Rmean = mean_[0],
Gmean = mean_[1],
Bmean = mean_[2];
double RGBmean = (Rmean + Gmean + Bmean) / 3;
double[] max_ = rgbHist.getHighValue();
double Rmax = max_[0],
Gmax = max_[1],
Bmax = max_[2];
double RGBmax = max(max(Rmax, Gmax), Bmax);
double fr = RGBmean / Rmean;
double fg = RGBmean / Bmean;
double fb = RGBmean / Bmean;
if (Double.isNaN(fr)) {
fr = 1;
}
if (Double.isNaN(fg)) {
fg = 1;
}
if (Double.isNaN(fb)) {
fb = 1;
}
double[] matrix = {//
fr, 0, 0,//
0, fg, 0,//
0, 0, fb};
return matrix;
}
/**
* Computes the white balance of an image using the Retinex algorithm.
* <p>
* References:<br>
* Edmund Lam, Combining gray world and retinex theory for automatic white
* balance in digital photography, Consumer Electronics, 2005. (ISCE 2005).
* Proceedings of the Ninth International Symposium on (2005), pp.134–139.
* <p>
* Kenfack, Pierre Marie. (2009). Implementierung und Vergleich verschiedener
* Algorithmen zur Bildsensorkalibrierung. Fraunhofer ITWM.
* http://www.itwm.fraunhofer.de/fileadmin/ITWM
* <p>
*
*
* @param rgbHist
* @return a 3x3 matrix which performs the color correction matrix*[R,G,B].
*/
public static double[] whiteBalanceRetinex(Histogram rgbHist) {
double[] mean_ = rgbHist.getMean();
double Rmean = mean_[0],
Gmean = mean_[1],
Bmean = mean_[2];
double RGBmean = (Rmean + Gmean + Bmean) / 3;
double[] max_ = rgbHist.getHighValue();
double Rmax = max_[0],
Gmax = max_[1],
Bmax = max_[2];
double RGBmax = max(max(Rmax, Gmax), Bmax);
double Rgain = Gmax / Rmax;
double Bgain = Gmax / Bmax;
if (Double.isNaN(Rgain)) {
Rgain = 1;
}
if (Double.isNaN(Bgain)) {
Bgain = 1;
}
double[] matrix = {//
Rgain, 0, 0,//
0, 1, 0,//
0, 0, Bgain};
return matrix;
}
/**
* Computes the white balance of an image using the Quadratic Mapping (QM)
* algorithm. QM is a combination of the Greyworld and the Retinex algorithm.
* And usually gives better results.
* <p>
* References:<br>
* Edmund Lam, Combining gray world and retinex theory for automatic white
* balance in digital photography, Consumer Electronics, 2005. (ISCE 2005).
* Proceedings of the Ninth International Symposium on (2005), pp.134–139.
* <p>
* Kenfack, Pierre Marie. (2009). Implementierung und Vergleich verschiedener
* Algorithmen zur Bildsensorkalibrierung. Fraunhofer ITWM.
* http://www.itwm.fraunhofer.de/fileadmin/ITWM
* <p>
*
* Returns a vector with 4 values: Rmu, Rnu, Bmu, Bnu.
* These values can be put into a 3x6 matrix as shown below:
* <pre>
*
* [ R
* G
* B
* [R' [ Rnu 0 0 Rmu 0 0 R^2
* G' = 0 1 0 0 0 0 * G^2
* B'] 0 0 Bnu 0 0 Bmu ] B^2 ]
*
*
* @param rgbHist
* @return a vector with the values {Rmu, Rnu, Bmu, Bnu}.
*/
public static double[] whiteBalanceQM(Histogram rgbHist) {
double[] max_ = rgbHist.getHighValue();
double Rmax = max_[0],
Gmax = max_[1],
Bmax = max_[2];
double R2max = Rmax * Rmax;
double G2max = Gmax * Gmax;
double B2max = Bmax * Bmax;
double Rsum = 0, R2sum = 0;
double Gsum = 0;
double Bsum = 0, B2sum = 0;
int[] bins = rgbHist.getBins(0);
for (int i = 0; i < bins.length; i++) {
Rsum += bins[i] * i;
R2sum += bins[i] * i * i;
}
bins = rgbHist.getBins(1);
for (int i = 0; i < bins.length; i++) {
Gsum += bins[i] * i;
}
bins = rgbHist.getBins(2);
for (int i = 0; i < bins.length; i++) {
Bsum += bins[i] * i;
B2sum += bins[i] * i * i;
}
double[] Rmunu = LinearEquations.solve(R2sum, Rsum, R2max, Rmax, Gsum, Gmax);
double[] Bmunu = LinearEquations.solve(B2sum, Bsum, B2max, Bmax, Gsum, Gmax);
double[] vector = {
Rmunu[0], Rmunu[1], Bmunu[0], Bmunu[1]//
};
return vector;
}
public static double[] whiteBalanceQM(long[][] rgbBins) {
double Rmax = 0,
Gmax = 0,
Bmax = 0;
double Rsum = 0, R2sum = 0;
double Gsum = 0;
double Bsum = 0, B2sum = 0;
long[] bins = rgbBins[0];
for (int i = 0; i < bins.length; i++) {
Rsum += bins[i] * i;
R2sum += bins[i] * i * i;
if (bins[i] != 0) {
Rmax = i;
}
}
bins = rgbBins[1];
for (int i = 0; i < bins.length; i++) {
Gsum += bins[i] * i;
if (bins[i] != 0) {
Gmax = i;
}
}
bins = rgbBins[2];
for (int i = 0; i < bins.length; i++) {
Bsum += bins[i] * i;
B2sum += bins[i] * i * i;
if (bins[i] != 0) {
Bmax = i;
}
}
double R2max = Rmax * Rmax;
double G2max = Gmax * Gmax;
double B2max = Bmax * Bmax;
double[] Rmunu = LinearEquations.solve(R2sum, Rsum, R2max, Rmax, Gsum, Gmax);
double[] Bmunu = LinearEquations.solve(B2sum, Bsum, B2max, Bmax, Gsum, Gmax);
double[] vector = {
Rmunu[0], Rmunu[1], Bmunu[0], Bmunu[1]//
};
return vector;
}
}