BLOCKFUN applies function on sub blocks of an array

partition an ND-array into sublocks and then applies a function on each block

Description

BLOCKFUN applies a function on blocks of an array
  Y=BLOCKFUN(X,S,funHandle) applies the function funHandle on blocks of size
  S on the array X. Y is of size ceil(size(X)./S)
  
  For instance, if X is of size [9 9] and S is [3 3], then
  X is partitionned in 9 [3 3] blocks as follow :
   | B1 | B4 | B7 |
   | B2 | B5 | B8 | where Bi's are [3 by 3] matrices
   | B3 | B6 | B9 |
  Y is then a [3 by 3] matrix with
   Y(i) = fun(Bi(:))
  
  This function is just a simple wrap for accumarray !
  See Also : accumarray
 
  EXAMPLES :
  rand('twister',12);
  x=floor(2*rand(9,9))
    x =
       0 0 1 1 1 0 0 0 0
       1 0 0 0 1 0 0 1 1
       0 1 1 1 0 0 0 0 0
       1 1 0 1 0 0 1 1 1
       0 1 0 0 0 0 0 0 0
       1 0 0 0 1 0 0 1 1
       1 1 0 1 0 1 1 1 0
       0 1 1 1 1 1 1 1 1
       1 0 0 1 0 0 0 0 1
 
  y=blockfun(x,[3 3],@sum)
  y =
       4 4 2
       4 2 5
       5 6 6
  
  y=blockfun(x,[3 3],@median)
  y =
       0 0 0
       0 0 1
       1 1 1
 
  y=blockfun(x,[5 5],@sum)
  y =
      12 5
      11 10
  
  x=floor(2*rand(6,6,6))
  y=blockfun(x,[3 2 3],@(x) length(find(x>0))>4)
 
  x=floor(2*rand(512,512,50));
  tic; y=blockfun(x,[5 5 5],@sum); toc

• MATLAB release: MATLAB 7.4 (R2007a)

[urs (us) schwarz]

good help with representative examples, nice error checks using the newly introduced ASSERTion routine, and an interesting, sleek computational engine WOULD make this contribution a very nice add-on - IF it wasn't for the enormous(!) amount of memory it needs during its intermediary steps (always of type double, even if the input is of a smaller data type), which - most unfortunately - makes this function almost unusable for even moderately big real-world problems, eg, a 3d set of mri scans of size
     512x512x20 (uint16)

us

[Mathias Ortner]

I updated the file to take care of Us comments.
The author