Recolor an image tile-wise with average colors of these certain tiles

Hello everyone,

I have the following image (original image is .tif format)

Now, I would like to recolor it, that I don't have single color pixels anymore. I imagine something like the following:

assess a 10 x 10 pixel wide tile (or any other size), add only the non-zero color values and devide it by the number of non-zero values in this 10 x 10 tile to get the average pixel color. Than, recolor this 10 x 10 tile with the average pixel color. And than repeat this, with all possible not overlapping tiles of the image.

I hope, someone is 1. abler to understand waht I tried to say and 2. also able to help me.

Thank you!

Julian

Answer 1

[Sean de Wolski]

Use blockproc to handle the block processing

doc blockproc

Then write a function which does the average ignoring zeros. So something like:

blockproc(magic(100),[10 10],  @(bs)(sum(bs.data(logical(bs.data)))./nnz(bs.data))*ones(size(bs.data)))

[Julian]

Perfect! That is almost exactly what I was looking for (except that it only returns grayscale videos).

Thank you!

[Sean de Wolski]

You could use it in a for-loop over each colorplane

szI = size(I);
I2 = zeros(szI,'like',I);
for ii = 1:szI(3)
   I2(:,:,ii) = what_I_Have(I(:,:,ii));
end

[Julian]

Great! It's always amazing, how fast someone is able to help around here :)

Thank you again!

Answer 2

[Matt J]

Splitting the image into tiles is easy using FEX: mat2tiles.

>> Image=rand(40,40); mat2tiles(Image,[10,10])

ans = 

    [10x10 double]    [10x10 double]    [10x10 double]    [10x10 double]
    [10x10 double]    [10x10 double]    [10x10 double]    [10x10 double]
    [10x10 double]    [10x10 double]    [10x10 double]    [10x10 double]
    [10x10 double]    [10x10 double]    [10x10 double]    [10x10 double]

You can then do all the operations you're talking about on each cell using a simple for-loop.