Finding pixel numbers for certain colors

Hi,

I want to be able to find which pixels are red which pixels are blue but with an extra issue. If a pixel is at the margin between red and blue, I want to find those pixels and group those pixels with any pixel that is two pixels away from it. I am attaching both the image I am working on and the one explaining the pixel issue. For the circled pixel in the image 'pixel selection', I need the pixel numbers of other marked ones. In the end since there will be coinciding pixel numbers (when it is applied for all pixels at the red-blue interface) I will eliminate the repeating ones directly. In the end I only need the pixel numbers for red, blue and red-blue interface.

Thanks in advance.

[Image Analyst]

Not sure of your definition of "pixel numbers" or "coinciding pixel numbers". Do you want the RGB histogram ( a count of the numbers of pixels with each unique color)?

[DGM]

The goal as described seems to be a list of the indices (or subscripts) of the 24 pixels surrounding each pixel in the transition area. That is, for subscripts, a big 600x1000x24x2 array (or some equivalent shape). How that would be used, I don't know.

That seems pretty questionable to me, so I'm just waiting to see if and why the direct use of the masks won't suffice.

Answer 1

[DGM]

That depends entirely on how you want to make the distinction between "red" or "blue" and colors which are in-between, because there are probably more colors in-between than what you think there are.

% the regions you think are uniform red/blue 
% are not uniform at all since this is a JPG
inpict = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/1480206/fullarea.jpg');
CT = [254 0 0; 0 0 254]; % the two color classes

% create masks
redmask = all(inpict == permute(CT(1,:),[1 3 2]),3);
blumask = all(inpict == permute(CT(2,:),[1 3 2]),3);

% the transition area is probably a lot larger 
% and more irregular than you expect because of that damage
F = imfuse(redmask,blumask); % visualze both masks together
imshow(F,'border','tight')

% generate transition mask
rbmask = ~redmask & ~blumask; % these are the pixels in-between
rbmask = imdilate(rbmask,ones(5)); % pixels within 2px of in-between pixels
imshow(rbmask,'border','tight')

That's probably not very useful. Since we need something other than an exact match, it's probably easier to approach this from hue alone.

inpict = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/1480206/fullarea.jpg');

% start by discriminating in H
% rotate H to avoid dealing with a split histogram
[H,~,~] = rgb2hsv(inpict);
H = mod(H-2/3+0.1,1);

% generate red/blue masks
H0 = [0.1 0.43]; % the target hues (shifted)
halfw = 0.05; % the tolerance
redmask = abs(H-H0(2)) <= halfw;
blumask = abs(H-H0(1)) <= halfw;

F = imfuse(redmask,blumask);
imshow(F,'border','tight')

% generate transition mask
rbmask = ~redmask & ~blumask;
rbmask = imdilate(rbmask,ones(5));
imshow(rbmask,'border','tight')

Note that the dilated rbmask overlaps both redmask and blumask now. If you need to remove the overlap area from redmask and blumask, just do

redmask = redmask & ~rbmask;
blumask = blumask & ~rbmask;

Similarly, the same thing could be accomplished by eroding those masks prior to generating rbmask, thereby eliminating the need to dilate rbmask.