How can i make a 2D convolution with x = [1 0 2; -1 3 1; -2 4 0] and h = [1 -1; 0 2] and not use conv2? Can you please help me.

I wanted to write a 2D convolution without conv2 where x = [1 0 2; -1 3 1; -2 4 0] and h = [1 -1; 0 2] but i get the wrong result of y=[7 0; 4 2] with the following code, can you help me?

m = 0:.1:pi;
x = [1 0 2; -1 3 1; -2 4 0];
h = [1 -1; 0 2]; 
res = conv2(x,h,'valid');
y = zeros(size(res));
for ii = 1:size(x,1)-size(h,1)+1
    for jj = 1:size(x,2)-size(h,2)+1
        y(ii,jj) = sum(sum(h.*x(ii:ii+size(h,1)-1,jj:jj+size(h,2)-1)));
    end
end

Answer 1

[Jan]

Look at the definition of the convolution again.

Either the orientation of h or the submatrices of x must be changed:

[h1, h2] = size(h);
[x1, x2] = size(x)

% Either:
h = rot90(h, 2);
for ii = 1:x1 - h1 + 1
    for jj = 1:x2 - h2 + 1
        y(ii,jj) = sum(h .* x(ii:ii+h1-1, jj:jj+h2-1), 'all');
    end
end

% Or:
for ii = 1:x1 - h1 + 1
    for jj = 1:x2 - h2 + 1
        y(ii,jj) = sum(h .* x(ii+h1-1:-1:ii, jj+h2-1:-1:jj), 'all');
    end
end