Is there any way to make this code faster?

Hi, I have this code that tosses out pixels from a cc list

    for i = 1:cc.NumObjects
        for j = length(cc.PixelIdxList{i}):-1:1
            ind = cc.PixelIdxList{i}(j);
            if ~imdiff(ind) || ~imcombined(ind)
                cc.PixelIdxList{i}(j) = [];
            end
        end
    end

imdiff and imcombined are both images. The code at the moment takes quite a while to churn through a decent sized image. Is there any way to make this code run faster?

Answer 1

[Daniel Shub]

I think this is a memory allocation issue. Every time you set cc.PixelIdxList{i}(j) = [], MATLAB needs to allocate new memory. You might be better off building an array that holds the indices to remove, and then removing them at the end.

[Qingyang]

Thanks! I tried what you suggested and it reduced the execution time by a ton! Here's the revised code:

      for i = 1:cc.NumObjects
          indices = false(length(cc.PixelIdxList{i}));
          for j = 1:length(cc.PixelIdxList{i})
              ind = cc.PixelIdxList{i}(j);
              if ~imdiff(ind) || ~imcombined(ind)
                  indices(j) = true;
              end
          end
          cc.PixelIdxList{i}(indices) = [];
      end

This works great but if you have any other suggestions to make it run even faster, please tell!

Answer 2

[Sean de Wolski]

That whole thing should be vectorizable assuming imdiff and imcombined (whatever those are) can handle it:

cellfun(@(c)c(imdiff(c)&imcombined(c)),CC.PixelIdxList,'uni',false)

[C.J. Harris]

Hmm.. is there any point using cellfun here? Does it increase speed or readability? At least in 2006b I find very few circumstances where cellfun gives improved performance over a simple loop.

[Sean de Wolski]

No. I just like it :)

You could keep the outer for-loop just as easily. The point is that the inner for-loop does not need to check each value if imdiff and imcombined are either images or functions that can be passed vector inputs.

[Qingyang]

Hi, thanks for the reply. I'm not quite familiar with that type of coding, but it did not give what I intended when I used it in place of the original code. I'm sure its probably just a small tweak that I'm can't tell. But, thanks anyways and I've got the code working much faster based on a suggestion from another commenter.