Hi, How to segment(divide) an image into four halves of "different dimensions" automatically as shown in the attached image using MATLAB?

As shown in the attached image, the 4 images in a 2×2 grid are of different dimensions. I want to extract them automatically into four different subplots. Now in this case all(4) images are not equally spaced still can we extract them automatically(not using hard coding) in Matlab? If they are equally spaced(same dimension) then it's easy to extract them "automatically", but I found difficulty in extracting them when these images are not equally spaced in a 2*2 grid. Need help in this regard. Thanks in advance.

[Zaidi Shoaib]

@Image Analyst

Answer 1

[DGM]

Here's one idea.

inpict = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/1234097/cars.jpg');
graypict = rgb2gray(inpict);

% filter the image with a wide difference filter
% this emphasizes long grid-aligned edges
fk = repmat([1; -1],[1 30]);
filtered = abs(imfilter(im2double(graypict),fk,'replicate'));
[~,rowidx] = max(mean(filtered,2)); % find the peak index
imshow(filtered,[])

% do the same thing with the filter rotated 90 deg
filtered = abs(imfilter(im2double(graypict),fk.','replicate'));
[~,colidx] = max(mean(filtered,1)); % find the peak index
imshow(filtered,[])

% disassemble the image using the peak locations
imgpile = cell(2,2);
imgpile{1,1} = inpict(1:rowidx,1:colidx,:);
imgpile{1,2} = inpict(1:rowidx,colidx+1:end,:);
imgpile{2,1} = inpict(rowidx+1:end,1:colidx,:);
imgpile{2,2} = inpict(rowidx+1:end,colidx+1:end,:);

% show the result
montage(imgpile,'size',[2 2])

Of course, this will have problems if the edges between subimages are weak or if there are more dominant grid-aligned edges in the composite image.

[Zaidi Shoaib]

@DGM thanks a lot for your effort. But it still involve some complexity.

Can you tell that how can we locate the common point of 4 images of 2*2 grid automatically? i think if we find that point then its computation become more easy.I Mention that Point in Attach File.what you think about this idea?can we locate that point automatically?

or second idea is that, can we detect edges 1st and then give that information to some command to separate and extract that portion?

[DGM]

That's not really considerably different. You're just finding two edges at once. I'm not so sure that it would have any significant advantage. I guess you could.

inpict = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/1234097/cars.jpg');
graypict = rgb2gray(inpict);

% construct the filter (center overlaps, but eh)
s = 30;
fk0 = repmat([1; -1],[1 s]);
fk = zeros(s);
fk(s/2+[0 1],:) = fk0;
fk(:,s/2+[0 1]) = fk0.';

% filter the image with a wide difference filter
% this emphasizes long grid-aligned edges
filtered = abs(imfilter(im2double(graypict),fk,'replicate'));
[~,rowidx] = max(mean(filtered,2)); % find the peak index
[~,colidx] = max(mean(filtered,1)); % find the peak index
imshow(filtered,[])

% disassemble the image using the peak locations
imgpile = cell(2,2);
imgpile{1,1} = inpict(1:rowidx,1:colidx,:);
imgpile{1,2} = inpict(1:rowidx,colidx+1:end,:);
imgpile{2,1} = inpict(rowidx+1:end,1:colidx,:);
imgpile{2,2} = inpict(rowidx+1:end,colidx+1:end,:);

% show the result
montage(imgpile,'size',[2 2])

[Zaidi Shoaib]

thanks @DGM.