BLOCKFUN.M

Execute a function on a sliding 2D window over a matrix

Syntax
OUT = blockfun( DATA, BLOCK, FUN )
OUT = blockfun( DATA, BLOCK, FUN, 'Name', Value, ... )
Description
OUT = blockfun( DATA, BLOCK, FUN ) returns a matrix OUT of dimensions
size(DATA), where each element corresponds to the value obtained by
exectuing FUN on the values within a 2D window defined by BLOCK centered
at each element in DATA.
BLOCK can be a user defined binary matrix, used as a mask centered at
each position to identify the elements passed to FUN. Or BLOCK can be a
1- or 2-element vector indicating the dimensions of the 2D window.
FUN is a function handle.

OUT = blockfun( DATA, BLOCK, FUN, 'Name', Value, ... ) accepts additional
name-value pairs from the following list (defaults in {}):
'includeCenter' { true } - if false, the center value of the 2D window
is excluded from the input to FUN.
'padFunction' { @nan } - a function handle used to pad the matrix DATA
when the 2D window extends beyond the bounds of the matrix.
'positions' { 1 : numel(data) } - a vector containing the linear indices
at which to execute FUN. When specified, OUT is of the same size as the
position vector.
'maxPosLength' { 1e4 } - a scalar indicating the maximum number of
positions to evaluate at a time. If this value is too large, MATLAB may
experience out of memory errors.

Examples
Determine the underlying shape of a surface with noise
[xx, yy] = meshgrid((1:100) - 51);

subplot(1,3,1);
surface = 2*xx + 0.4*yy + 0.2*xx.*yy;
imagesc(surface); axis square; axis off;

subplot(1,3,2);
noisy = surface + randn(size(surface))*100;
imagesc(noisy); axis square; axis off;

subplot(1,3,3);
recovered = blockfun( noisy, [6 6], @nanmedian );
imagesc(recovered); axis square; axis off;