One last (??) version that is completely vectorized.  Previous versions looped over the number of neighbors.

Tony

(file #49519)

I have attached code for a function I've called colfilt2.  It is similar to colfilt but allows you to process only neighborhoods of selected pixels. It also allows you to choose members of the neighborhood by specifying row and column index offsets instead of using all members of a rectangular region.

Like colfilt, colfilt2 requires that a processing function be passed as an argument. The processing function must accept a matrix argument. In the processing function for colfilt each column of the matrix contains the neighbors in a rectangular region around an element of the input image. The number of columns is equal to the number of pixels in the image.  In the processing function for colfilt2 each column contains only the specified neighbors of a pixel.  The number of columns is equal to the number of selected pixels.  This processing function is only called once (unlike in nlfilter where the processing function is called once for each pixel in the input image).

Here is example code that illustrates usage:

%% Replace diagonal elements with average of left and right
%% neighbors. Zero-padding is used at the edges
A = magic (12)
SELECT = eye(size(A), 'logical');
NEIGHBORS = [0 -1; 0 1];
colfilt2 (A, SELECT, NEIGHBORS, @mean)

This example is included as "demo" code in the function. It is  a modified version of the demo code in colfilt.

Tony Richardson