How to make an image take different RGB values?

I have an image with NxMx3 uint8 colors but I want the colors of the image to be replaced with another array of 256x3(Each column corresponds to R,G,B values) colors. How can I map the values from the array to the image?

I was advised that I need to split the 3 channels and then map them independently

Answer 1

[Jan]

LUT = randi([0, 255], 256, 3, 'uint8');       % The look up table
Img = randi([0, 255], 640, 480, 3, 'uint8');  % The image

NewImg = cat(3, LUT(Img(:, :, 1) + 1, 1), ...
                LUT(Img(:, :, 2) + 1, 2), ...  % [EDITED, 3 ==> 2]
                LUT(Img(:, :, 3) + 1, 3));

This replaces the original colors by the ones from the look up table by a simple indexing. See this example:

LUT = [2, 4, 8]
Img = [0, 1; 2, 1];
LUT(Img + 1)

[Abdul Rahim Mohammad]

Wouldnt

NewImg = cat(3, LUT(Img(:, :, 1) + 1, 1), ...

LUT(Img(:, :, 2) + 1, 3) wouldn't this be mapping lookuptable column 2 to column 3 of the image?), am I correct in assuming that its supposed to be 2 instead of 3

LUT(Img(:, :, 3) + 1, 3));

Also, would this code:
RGB_map is a 256x3 array
R_map = RGB_map(:,1); %Red channel
G_map = RGB_map(:,2);
B_map = RGB_map(:,3);

newimage = cat(3, (R_map+1,1), (G_map+1,2), (B_map+1,3));

perform the same considering R,G,B_map are 256x1 arrays derived from a RGB_map 256x3 array

[Guillaume]

Yes, Jan made a typo. That 2 should be 3.

No your code is not equivalent. You don't use the input image anywhere.

%your RGB_map is Jan's LUT variable
R_map = RGB_map(:,1); %Red channel
G_map = RGB_map(:,2);
B_map = RGB_map(:,3);
%Img is the input image

newimage = cat(3, R_map(Img(:, :, 1) + 1, G_map(Img(:, :, 2) + 1, B_map(Img(:, :, 3) + 1))))

would be equivalent.

[Abdul Rahim Mohammad]

Thank you both very much, this has been very helpful.

[Abdul Rahim Mohammad]

So, I have yet to get this actually working on my code, here's the problem:

Given an rgb image nxmx3 and the 256x3 map . For each pixel I want to look up the red value in the red column of the 256x3 table, the green in the green column and the blue in the blue column. This should get me a rgb value that I can look up in the map and it should tell me what the corrected rgb value should be. I want to save this set of rgb values in the output image and repeat the process for all other pixels in the image.

How would my code perform:

LUT = RGB_map;    % look up table
Img = RGB_image;  % passed image
[r c] = size(Img);
image_Rmap = Img(:,:,1);
image_Gmap = Img(:,:,2);
image_Bmap = Img(:,:,3);
R_map = RGB_map(:,1); %Red channel
G_map = RGB_map(:,2);
B_map = RGB_map(:,3);
pixel = zeros(r,c);
for i = 1:r
    for j =1:c
        pixel = [image_Rmap, image_Bmap, image_Gmap];
    end
end

[Guillaume]

I'm unclear why Jan's solution doesn't match what you describe. Can you explain with a an example what you want exactly.

e.g: given an image pixel with RGB values

pixel = uint8([0 127 255]);

and a colour map

LUT = uint8([(0:255)', (255:-1:0)', mod(-128:127, 256)']);

what pixel value do you want as output?

[Abdul Rahim Mohammad]

Let me explain my problem more clearly, I have a image with test code that i run through the to make see how my outputs, This particular problem is a part of a much bigger problem:

This is my test code:

% A simple TEST for colorchart code


truecard = imread('color_testchart.jpg');
RGB_truechart = get_chart_values(truecard);

trueimage = imread('color_toys.jpg');  

% test example 1 - reduce all red tones by 50% (cyan colored)
badimage = trueimage;
badimage(:,:,1) = 0.5*badimage(:,:,1);
badtestcard = truecard;
badtestcard(:,:,1) = 0.5*badtestcard(:,:,1);

RGB_badchart = get_chart_values(badtestcard);
RGB_map = chart_correction(RGB_truechart,RGB_badchart);
fixedcard = apply_rgb_map(badtestcard,RGB_map);
fixedimage = apply_rgb_map(badimage,RGB_map);

figure
subplot(2,3,1); image(truecard);title('Reference Card');
subplot(2,3,2); image(badtestcard);title('Bad Card');
subplot(2,3,3); image(fixedcard);title('Fixed Card');
subplot(2,3,4); image(trueimage);title('Reference Image');
subplot(2,3,5); image(badimage);title('Bad Image');
subplot(2,3,6); image(fixedimage);title('Fixed Image');

The Functions get_chart_values extracts the values from a chart image and output RGB_list = 24x3 table

The function chart_correction takes the extracted values from get_chart_values 24x3 and bad_chart(supplied in the test code), both of these are combined to form a 256x3 table of colors for each of the 24 colors in get_chart_values using polyfit and polyval

The function apply_rgb_map takes 256x3 output of chart_correction and applies the corrected colors to entire image.

When I apply this code for the rgb_map, the outputs of the images go to from 200000x1x3 and I'm not sure why that happens.

[Jan]

This code:

for i = 1:r
    for j =1:c
        pixel = [image_Rmap, image_Bmap, image_Gmap];
    end
end

is meaningless, because the value of pixel is overwritten in each iteration, so you can simply omit the loop. You would need:

for i = 1:r
    for j =1:c
        pixel(i, j, :) = [Rmap(image_Rmap(i,j), ...
                          Bmap(image_Bmap(i,j), ...
                          Gmap(image_Gmap(i,j))];
    end
end

But the code I have posted does this much faster, because it is vectorized.

"When I apply this code for the rgb_map"

When you apply which code? I cannot guess this detail. Do you mean the output of:

NewImg = cat(3, LUT(Img(:, :, 1) + 1, 1), ...
                LUT(Img(:, :, 2) + 1, 2), ...  % [EDITED, 3 ==> 2]
                LUT(Img(:, :, 3) + 1, 3));

? Then add a reshape:

NewImag = resize(NewImg, size(Img));

I'm still convinced, that my code solves the question.

What is the answer to Gauillaume's question (link)?

[Guillaume]

Jan made a typo. Replace resize by reshape and all will be well.

[Abdul Rahim Mohammad]

Edit: I got it working, NewImag = reshape(NewImg,size(img)); works to reshape the array into my desired dimensions.