Asked by Itzik Ben Shabat
on 27 Apr 2013

Hi, suppose I have a matrix M which is nXn size. i wish to turn this matrix into a smaller matrix M1 which is n/numXn/num size. each element of M1 will contain the sum of numXnum elements of the original M. for example: M1(1,1)=sum(M(1:num,1:num))

is there a function in matlab that does this in an efficient way or am i forced to use a for loop and handle the case where n/num is not an integer? thanks

*No products are associated with this question.*

Answer by Cedric Wannaz
on 27 Apr 2013

Edited by Cedric Wannaz
on 27 Apr 2013

Accepted answer

There is no function, up to my knowledge, that does this automatically (as you can see here, I already posted a similar question [in a different context] that was not answered). It is, however, not too complicated to build a little piece of code for this.

So, If I understand well, you want to compute some block-statistics on `M (nxn)`, using a block size that might not be an integer divider of `n`.

Let's discuss a simple case:

>> n = 10 ; >> bSize = 3 ; % 3x3 blocks. >> M = randi(30, 10, 10) % Rand int M for the example. M = 25 5 20 22 14 9 23 26 11 3 28 30 2 1 12 21 8 8 25 2 4 29 26 9 23 20 16 25 18 16 28 15 29 2 24 5 21 8 17 24 19 25 21 3 6 4 27 28 28 29 3 5 23 25 15 15 29 11 9 4 9 13 23 21 14 29 17 6 23 18 17 28 12 10 20 11 5 8 23 15 29 24 20 29 22 18 5 19 12 1 29 29 6 2 23 7 8 15 18 11

The first thing to note is that there are multiple ways to split 10 in blocks of 3 or almost 3.

3, 3, 3, 1 3, 3, 2, 2 3, 2, 3, 2 4, 3, 3 etc

and you'll want one or the other way depending on whether you need to have as many 3x3 blocks as possible, or if you want to balance the number of elements in each block.

So you have to choose how you want to define block sizes along both dimensions (you can CIRCSHIFT them to make the distribution of block sizes a little more homogeneous). I'll choose `4,3,3` along both dimensions for the example.

Now one way to achieve what you want is to build a matrix of block IDs, that we use then to summarize `M` on blocks, i.e

1 1 1 1 2 2 2 3 3 3 1 1 1 1 2 2 2 3 3 3 1 1 1 1 2 2 2 3 3 3 1 1 1 1 2 2 2 3 3 3 4 4 4 4 ... 4 4 4 4 ... 4 4 4 4 ... 7 7 7 7 ... 7 7 7 7 ... 7 7 7 7 ...

We can build such a matrix as follows:

>> base = floor((0:n-1) * bSize/n) base = 0 0 0 0 1 1 1 2 2 2

>> [JJ, II] = meshgrid(base, bSize*base) ; >> blockIDs = 1 + II + JJ blockIDs = 1 1 1 1 2 2 2 3 3 3 1 1 1 1 2 2 2 3 3 3 1 1 1 1 2 2 2 3 3 3 1 1 1 1 2 2 2 3 3 3 4 4 4 4 5 5 5 6 6 6 4 4 4 4 5 5 5 6 6 6 4 4 4 4 5 5 5 6 6 6 7 7 7 7 8 8 8 9 9 9 7 7 7 7 8 8 8 9 9 9 7 7 7 7 8 8 8 9 9 9

Now we can obtain any block-stat. using ACCUMARRAY, where we use `M` as values, and `blockIDs` as indices. For the sum and the mean per block, for example:

>> blockSum = accumarray(blockIDs(:), M(:)) blockSum = 275 196 183 190 156 156 235 119 122

>> blockMean = accumarray(blockIDs(:), M(:), [], @mean) blockMean = 17.1875 16.3333 15.2500 15.8333 17.3333 17.3333 19.5833 13.2222 13.5556

Hope it helps!

Opportunities for recent engineering grads.

## 0 Comments