Matrix Standard Deviation in 3D neighborhood

Hi

I have a 3D matrix (NxNxM) and I want to compute standard deviation in a sliding window of, say, 3x3xM which means a square window going across all the matrices (or slices) but sliding in 2D only? Can anyone help me on this.

stdfilt() does not do what I want. It computes a 3x3x(odd number) window standard deviation which is perhaps sliding in depth or 3rd dimension as well.

Best Regards

Wajahat

Answer 1

[Andrei Bobrov]

% I - your 3D matrix - array(NxNxM)
k = padarray(I,[1 1],'symmetric');
s = size(I);
out = zeros(s(1),s(2));
for j1 = 1:size(I,1)
    for j2 = 1:size(I,2)
        out(j1,j2) = std2(k(j1 + (0:2),j2 + (0:2),:));
    end
end

Answer 2

[Sean de Wolski]

stdfilt() will ignore the three-dimensional component if you tell it to:

x3 = stdfilt(repmat(magic(10),[1 1 3]),ones(3,3,1)) %three-d array, all slices equal
x1 = stdfilt(magic(10),ones(3,3,1)) %two-d array

all equal?

all(all(all(bsxfun(@eq,x3,x1))))

[Wajahat]

I do not want to ignore the three dimensional component, I want to include it but I want a 2D array at the end.

As I have understood after the two answers, x3=stdfilt(repmat(magic(10),[1 1 3]), ones(3,3,5));

x3=x3(:,:,1);

will also do the same thing as Andrei Bobrov answer above.

With a window of 5 in depth, stdfilt() will center the window on the first slice and will include all the slices of 3-Slice 3D array while having a 3x3 window in 2D. Window size in depth will be limited by the number of available slices.

The documentation for stdfilt() should also explain how the selected neighborhood or window is placed in 3D.