Multiply Image Pixels by a Matrix

Hello, Let's say I have a 3 channel image M -> [numRowx * numColumns * numChannels]. I also have a 3 by 3 matrix A (Color Conbersion).

I would like to multiply each pixel 3 channel by the matrix to have a new 3 channels pixel.

How would you do that the most efficient way without writing the actual formula by hand (Multiplying each channel of the image by a row)?

My best so far is:

rgbToYPbPrMat = [0.299, 0.587, 0.144; 0.5, -0.419, -0.081; -0.169, -0.331, 0.5];
yPbPrToRgbMat = inv(rgbToYPbPrMat);

inputImage = randi([0, 255], [256, 256, 3], 'uint8');
inputImageFP = double(inputImage) / 255; %<! Floating Point

figure;
image(inputImage);
axis image;

inputImageYPbPrFP = shiftdim(reshape((rgbToYPbPrMat * reshape(shiftdim(inputImageFP, 2), [3, (256 * 256)])), [3, 256, 256]), 1);

Thanks.

[Sean de Wolski]

I smell bsxfun. Please provide a small example with a sample image we have and a a 3x3 matrix. What would you expect as the result?

[Royi Avital]

Hi @Sean, Look at my code I added.

I couldn't make it work with 'bsxfun'.

[Royi Avital]

Anyone? Please...

Answer 1

[Jan]

I cannot check it currently, but please try:

M = rand(numRows, numColumns, 3);
T = rand(3, 3);
P = reshape(M, numRows * numColumns, 3) * T';
Result = reshape(P, numRows, numColumns, 3);

[Royi Avital]

Hi, Checked it, it does work and it the fastest so far (Since the Transpose works on the smaller matrix).

Do you think there's even a faster way?

Thanks.