You’d have to define red to begin with. Is it only [1 0 0] or are there other choices? Once you have the choices, you can then apply logic to see which pixels are red.

–Loren

Thanks

Alex

I just did a quick search, and found that accumarray was first introduced in R14, and then the interesting upgrades to it came with R14SP3. I first used accumarray with a version prior to that, which is when I had issues with it. I’m pretty sure my university skipped R14SP3, and by the point they upgraded to a later release, I no longer depended on accumarray enough to notice the improvements.

It seems like a fantastically versatile function but for some reason I’ve found it quite difficult to wrap my head around it. These examples help quite a bit. Thanks again!

-Wes

subs = [1 1; 2 1; 2 3; 2 1; 2 3];
A = accumarray(subs, 101:105, [2 4], @(x)length(x)>1)

subs = [1 1; 2 1; 2 3; 2 1; 2 3];
A = accumarray(subs, 101:105, [2 4], @(x){x})

go even further.

You are right, that color counting example does benefit from using the 1-D form of the indexing arg (even Loren, who, believe me, knows a lot about MATLAB, misses simple things sometimes). It’s faster, though perhaps not for the reason you were thinking: it’s the creation of the 2-column indexing argument that seems to take the bulk of the extra time in the cases I looked at.

And, as you probably noticed, specifying the size of the output (via the third input arg) saves ACCUMARRAY the trouble of figuring it out, also saving time.

- Peter

There is, however, one oddity about the example given here: Why did you use two dimensional indexing by concatenating with ones(M,1)? The transpose of the result from using one-dimensional indices (i.e., just X(:)) is equivalent, and for large images on my computer, seems to take only ~25% of the time:

>> tic; H = accumarray([ones(M,1) X(:)], 1 , [1 nc]); toc
Elapsed time is 0.483881 seconds.
>> tic; H2 = accumarray(X(:), 1, [nc 1])’; toc
Elapsed time is 0.127087 seconds.
>> isequal(H,H2)

ans =

1

Please look at the documentation in which there are examples for just your problem. You may also be interested in reading Steve’s blog on image processing.

–Loren

Just as an update, I ran this M-file as a script and as a function with some timings of the first counting problem and get the following results on my dual-core T60 laptop:

timesScript =
     for      sparse   accumarray
    0.2425    0.0478    0.0210

timesFunction =
     for      sparse   accumarray
     0.0063    0.0334    0.0175

In any case, there still remain ways to do accomplish the same goal, some of which may be better suited to your needs than others.

–Loren