How to convert m-by-n-by-3 array to m-by-n?

I am attempting to manually convert RGB values to HSI. I do not wish to use the rgb2hsv code included in MATLAB. Also, the hope is to manually recreate other color spaces I have found in research articles. Therefore, solving this problem would assist me immensly in those areas. Here is my code for the HUE component...

for i = 1:x
    for j = 1:y
      
        TH(i,j) = acos((((1/2).*((R(i,j) - G(i,j)) + (R(i,j) - B(i,j))))./(sqrt((R(i,j) - G(i,j)).^2 + ((R(i,j) - B(i,j)).*(G(i,j) - B(i,j)))))));
        if B(i,j) <= G(i,j)       
        H(i,j) = TH(i,j);
        elseif B(i,j) > G(i,j)
        H(i,j) = ((2*pi) - TH(i,j));
        end 
    end
end

...

The idea is that it will loop through each pixel and create THETA based on the formula, then assign a value to the HUE component image at that pixel based on the comparison of blue and green pixel values at that point (as shown in the code). My issue is that it does create H; however, H is an m-by-n-by-3 array, and i would like to have a m-by-n array. Any assistance in the matter would be appreciated and will serve as foundation for me to try an emulate other color spaces. Also if there is a cleaner way beside a for loop feel free to share.

THANK YOU

[Walter Roberson]

Your code before this point assigns something 3 dimensional to H.

Unless, that is, that instead of H being m by n by 3 like you appear to write, that you mean that H is coming out m by (n times 3) . If that is what is happening, then go back and re-read the documentation of size() for the case where you have fewer outputs than the number of dimensions in your array.

Also if there is a cleaner way beside a for loop

Logical indexing. Vectorize the calculation of TH and then create a mask of places B < G and assign to H(mask) from source TH(mask) . And so on.

[Johnathon Street]

yes, you were correct in that. I immediately found the issue after reading your comment. I was creating an array of H = zeros(size(im)), where im is the original rgb image. Another step in the right direction. Thank you. I will search logical indexing and attempt it. If it isnt too much trouble perhaps you could provide an example?.

[Walter Roberson]

x = randi(9, 1, 20);
mask = x > 5;
y = zeros(size(x));
y(mask) = x(mask) / 2;
mask = x <= 3;
y(mask) = x(mask) + 3;

[Johnathon Street]

im = Colorcloud;
im = rescale(im);

R = im(:,:,1);
G = im(:,:,2);
B = im(:,:,3);

TH = (acos((1/2).*((R - G) + (R - B)))./(sqrt((R - G).^2 + (R - B).*(G - B))));
mask = (B <= G);       
H(mask) = TH(mask);
mask = (B > G);
H(mask) = ((2*pi) - TH(mask));
H = (H./(2*pi));

Previous effort using different image was successful. However, I did not get the same H image as i had when running the for loop. The results were slightly different. When running with different image i recieve this error...

Attempt to grow array along ambiguous dimension.

Error in code (line 11)
H(mask) = TH(mask);
 

Do you see where i may have made an error in my code?

[Walter Roberson]

 H = TH;

Then skip the first test. The second test will overwrite all relevant parts of H.

If I recall correctly for computing hue you need tests involving different color channels not just B and G.