Referenced White Balance of an image

Hello,

I have a basic script for white balancing an image and it works quite okay. However now I am looking to modify it so it will correct a incorrectly balanced image by means of the analysis of a second image that will be used as white reference and have no idea where to even start.

This is the code I have and the effect:

% Read input image

img = imread('wb.jpg');

% Convert image to double precision

img = im2double(img);

% Get image dimensions

[rows, cols, channels] = size(img);

% Compute average channel values

Ravg = mean(mean(img(:,:,1)));

Gavg = mean(mean(img(:,:,2)));

Bavg = mean(mean(img(:,:,3)));

% Compute correction factors

Kavg = (Ravg + Gavg + Bavg) / 3;

Kr = Kavg / Ravg;

Kg = Kavg / Gavg;

Kb = Kavg / Bavg;

% Apply correction factors to each pixel

for i = 1:rows

for j = 1:cols

img(i,j,1) = Kr * img(i,j,1);

img(i,j,2) = Kg * img(i,j,2);

img(i,j,3) = Kb * img(i,j,3);

end

end

% Convert image back to uint8

img = im2uint8(img);

% Display original and corrected images

figure;

subplot(1,2,1); imshow('wb.jpg'); title('Original Image');

subplot(1,2,2); imshow(img); title('Corrected Image');

Now I am looking to modify the code so it takes both the image we are looking to white balance, and a second image that has the same lighting conditions as the first one.

The output should be a correctly white balanced image.

I will be thankful for any help.

Answer 1

[DGM]

I really have no familiarity with canonical white balancing techniques, but I'll shoot. First, let's simplify things

% Read input image
inpict = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/1368389/image.png');

% Convert image to double precision
inpict = im2double(inpict);

% Compute average channel values
RGBavg = mean(inpict,1:2);

% Compute correction factors
Kgray = mean(RGBavg);
Krgb = Kgray./RGBavg;

% Apply correction factors
outpict = inpict.*Krgb;

% Convert image back to uint8
outpict = im2uint8(outpict);

% Display original and corrected images
figure
subplot(1,2,1); imshow(inpict); title('Original Image');
subplot(1,2,2); imshow(outpict); title('Corrected Image');

At that point, I don't see why you can't just use the calculated correction factors.

% Read test image
testpict = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/1368394/image.png');

% Apply correction factors
outpict2 = im2double(testpict).*Krgb;

% Convert image back to uint8
outpict2 = im2uint8(outpict2);

% Display original and corrected images
figure
subplot(1,2,1); imshow(testpict); title('Original Image');
subplot(1,2,2); imshow(outpict2); title('Corrected Image');

[DGM]

If you're asking "why is the output blue?", that's because the given process is heavily skewed by image content which is not nominally-white. In order to make the image mean neutral in the presence of red/brown content, white regions become blue.

Consider restricting the sampling to regions which can be considered white.

% Read input image
inpict = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/1368389/image.png');

% Convert image to double precision
inpict = im2double(inpict);

% Convert to LAB and get thresholds
% these values split the distribution of L,C at their center
% for 50%, you could use mean(), but this can otherwise be generalized
[L A B] = imsplit(rgb2lab(inpict));
C = hypot(A,B);
Llimit = mean(stretchlim(L/100,0.49))*100;
Climit = mean(stretchlim(C/134,0.49))*134;

% create masks from L,C
lmask = L > Llimit; % brightest half of pixels
cmask = C < Climit; % least colorful half of pixels
whiteregion = repmat(lmask & cmask,[1 1 3]);

% Compute average channel values in selected region
thesepixels = reshape(inpict(whiteregion),[],1,3);
RGBavg = mean(thesepixels,1);

% Compute correction factors
Kgray = mean(RGBavg);
Krgb = Kgray./RGBavg;

% Apply correction factors
outpict = inpict.*Krgb;

% Convert image back to uint8
outpict = im2uint8(outpict);

% Display original and corrected image
% compare against the original global averaging method
op1 = imread('op1.png');
figure
imshow([im2uint8(inpict); op1; outpict],'border','tight')

% Read test image
testpict = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/1368394/image.png');

% Apply correction factors
outpict2 = im2double(testpict).*Krgb;

% Convert image back to uint8
outpict2 = im2uint8(outpict2);

% Display original and corrected image
% compare against the original global averaging method
op2 = imread('op2.png');
figure
imshow([im2uint8(testpict); op2; outpict2],'border','tight')

Answer 2

[Image Analyst]

This is the best algorithm I've run across:

Mark Grundland Color Histogram Specification by Histogram Warping

Sorry, I don't have code for it.