Can this be vectorised? (Adding 3D pattern to 3D matrix at specific indices)

I currently have a problem which requires adding a 3D pattern (stored as a 3D matrix) to another 3D array at a particular X,Y point, specified in a different 2D matrix of 1s and 0s. At a '1' I need to add the pattern on top of whats there, and do nothing at a '0'.

Here is the code:

for i=1:sizei
    for j=1:sizej
        if edge_image(i,j)==1
            range_i = i-shapesize/2:i+shapesize/2;
            range_j = j-shapesize/2:j+shapesize/2 ;
            voting_area(range_i,range_j,:)= voting_shape + voting_area(range_i,range_j,:) ; 
        end
    end
end

Edge_image is the binary 2d array, voting_shape is the 3d pattern matrix and voting_area is the bigger 3d matrix. (This is a Hough transform loop looking for rotations of a square in an edge image).

Currently this is extremely slow, but the fact that the 2d image is just a binary matrix makes me suspect there could be a way to speed this loop up. Is this possible at all?

Cheers

Answer 1

[Matt J]

 edge_vol=repmat(edge_image,1,1,size(voting_shape,3));

 voting_area= voting_area + convn(edge_vol,voting_shape,'valid');

[n m]

Perfect! thankyou

[Matt J]

I think maybe that should have been

 convn(edge_vol,voting_shape,'same')  %not 'valid'