Non-Overlapping Moving Sum

Consider any matrix, say Rain=(1:10)'. I want to compute a non-overlapping moving sum with two element.

Rain=1,2,3,4,5,6,7,8,9,10

Compute=1+2,3+4,5+6,7+8,9+10

With traditional movsum command, the moving sum is overlapping elements. I want to evaluate non-overlapping moving sum. Any in-built function? I have tried with loops and all those, it works but in-built function would be fast to compute.

Answer 1

[Star Strider]

Try this —

Rain = [1,2,3,4,5,6,7,8,9,10];
rRain = reshape(Rain, 2, [])

rRain = 2×5

1 3 5 7 9 2 4 6 8 10

sumRain = sum(rRain)

sumRain = 1×5

3 7 11 15 19

These could be combined into one line, however I kept them separate to demonstrate how it works.

.

[Suman Dhamala]

Actually my matrix has size 135*129*118. I wanted to accomplish this non-overlapping 2 sum so that my final matrix is 135*129*59. I tried using reshape, but may be i did some mistake the resultant matrix wasn't the one i wanted, could you provide some hint for 3-d matrix?

[Star Strider]

It always helps to have the actual problem in the beginning.

Every combination of permute, reshape and sum that I experimented with failed to provide the desired result.

.

[Image Analyst]

Star's right - avoids wasting time if we know the full situation in advance. However he showed you a nice trick that will come in useful. It's also in the FAQ:

https://matlab.fandom.com/wiki/FAQ#How_do_I_compute_the_mean_of_a_vector_in_blocks.3F

[Matt J]

You can download sepblockfun, which generalizes the technique Star Strider has shown to higher dimensions.

A=rand(135,129,118);
B=sepblockfun(A,[1,1,2],'sum');

whos A B

  Name        Size                    Bytes  Class     Attributes

  A         135x129x118            16439760  double              
  B         135x129x59              8219880  double              

[Star Strider]

@Matt J — Thank you! I didn’t think to look in FEX.

[Suman Dhamala]

This worked thanks @Matt J and everyone

Answer 2

[Image Analyst]

Since you have an image, you can do it with blockproc. However it only works with 2-D arrays so you'll have to do it once on each slice, then again along the z direction. I'm attaching some blockproc demos. I haven't done it with a 3-D image so you're on your own but I'm pretty sure it can be done.