convert larger image into patches

hi there,

I currently have a large image sized 600x1000 and would like to instead read this as patches of 51x51, how can I do this?

have tried to reshape but number of elements in each doesn't match to an integer

thanks very much

Answer 1

[Ameer Hamza]

You can use this FEX package: https://www.mathworks.com/matlabcentral/fileexchange/35085-mat2tiles-divide-array-into-equal-sized-sub-arrays

img = rand(600,1000);
img_parts = mat2tiles(img, [51 51])

[cameron lord]

Thankyou, I used mat2tiles and it splits the image up - but i am trying to multiply each of the patches with a 51x51 filter matrix but cannot as my original image is not a multiple of 51 so some of the matrix dimensions disagree.

is there a way to get round this?

[Ameer Hamza]

You can remove those cells

img = rand(600,1000);
img_parts = mat2tiles(img, [51 51]);

idx = ~cellfun(@(x) all(size(x)==[51 51]), img_parts);
img_parts(:,all(idx,1)) = [];
img_parts(all(idx,2),:) = [];

Answer 2

[Image Analyst]

What exactly do you mean by "read this"? Do you want to resize the entire image to 51x51 with imresize(yourImage, [51, 51]) or do you want to scan the image with a 51x51 window, moving in jumps of 51 pixels, and process the image inside, like you'd do with blockproc(). I'm attaching images for blockproc.

If the image is not a perfect multiple of 51, what do you want to do with the left over sliver?

[cameron lord]

yes, your right with the second one. i'd like to process the whole image - but in chunks of 51x51, passing them through a filter then plot the resulting activation map.

as for the sliver that will be left over, i'm not entirely sure. perhaps just remove it from the activation map completely?

[Image Analyst]

OK, so simply crop the image before blockproc() with indexing. Just compute the number of blocks.

numBlocksR = size(yourImage, 1);
numBlocksC = size(yourImage, 2);
yourImage = yourImage(1 : (floor(numBlocksR/51) * 51), 1 : (floor(numBlocksC/51) * 51), :);