How do I store a series of RGB images in a 2D array?

Hi,

I'm working on a CNN project and I want to tile an image 224x224x3 with a stride factor S=4 and block size of 11x11. This corresponds to an output 2D array of size 54x54, where each element would be a 11x11x3 image tile.

I have generated these tiles using a for loop. What I wish to do is have an output array of size 54x54, where each element holds an 11x11x3 image tile. So if I index output_array(1,1) it will give me the first 11x11x3 image array I have tiled using my for loop.

Is this possible with matlab, and how would I go by achieving this?

My pseudocode so far:

// load image

for i=1:54
   for j=1:54
      // tile image by stride and blocksize
      output_array(i,j)=tiled_image; //this does not work, I do not know how to initialize output_array
   end
end

Help appreciated!

[tekiwibird]

I just found out that output_array{i,j}=tiled_image would work, are there any downsides of using this?

[Rik]

There isn't really a downside to using cells, although indexing can be tricky sometimes. I believe you can remove the need for a for-loop with mat2cell.

Answer 1

[Guillaume]

You can indeed use a 2D cell array where each cell is one of the 3D matrix tile. This may be the simplest thing for you. You could also use a 5D matrix where each matrix5D(:, :, :, xtile, ytile) (or squeeze(matrix5D(xtile, ytile, :, :, :))) is one of the tile.

Processing will probably be faster with a 5D matrix than with the cell array, but I would recommend you use whichever form you're most comfortable with. Worry more about code clarity rather than speed to start with. Prologue for either case:

%img: source image
stride = [5, 5];       %height, width
blocksize = [11, 11];  %height, width
tilescount = (size(img(:, :, 1)) - blocksize - 1) / stride + 1;
assert(all(mod(tilescount, 1) == 0), 'cannot divide image into tile evenly')

Generating a cell array

tiles = cell(tilescount);
tileidx = 1;
for col = 1 : stride(2) : size(img, 2 ) - blocksize(2)
   for row = 1 : stride(1) : size(img, 1) - blocksize(1)
      tiles{tileidx) = img(row:row+stride(1)-1, col:col+stride(2)-1, :);
      tileidx = tileidx + 1;
   end
end

Generating a 5D matrix:

tiles = zeros([tilescount, blocksize, size(img, 3)]);
tilecol = 1;
for col = 1 : stride(2) : size(img, 2 ) - blocksize(2)
   tilerow = 1;
   for row = 1 : stride(1) : size(img, 1) - blocksize(1)
      tiles(tilerow, tilecol :, :, :) = img(row:row+stride(1)-1, col:col+stride(2)-1, :);
      tilerow = tilerow + 1;
   end
   tilecol = tilecol + 1;
end
%optionally afterward, if you prefer indexing with tiles(:, :, :, tilerow, tilecol)
%tiles = permute(tiles, [4 5 1 2 3]);