Problem trying to access labeled portion of images (updated 7/5)

UPDATE: data_dump.mat

unique(diff(unique(endpts_labeled)))
ans = 
     1
     2

ORIGINAL QUESTION:

I am getting a rare error running the following code on an image of 1px thick skeleton segments:

    V(V<=0) = 0; V(V>0) = 1; % Segments <=2 pixels are removed as well earlier
    [K n] = bwlabel(V,8);
    props = regionprops(K,'Perimeter');
    tort = zeros(1,n);
    endpts = bwmorph(K,'endpoints');
    endpts_labeled = immultiply(K,endpts);
    for m = 1:n
        [rows,cols] = find(endpts_labeled==m);
        p1 = [rows(1),cols(1)]; % Error happens here!!
        p2 = [rows(2),cols(2)];
        d = sqrt((p1(1)-p2(1))^2+(p1(2)-p2(2))^2);
        c = props(m,1).Perimeter/2;
        tort(m) = c/d;
    end

The error happens when the loop tries to access rows(1) when the numel(rows) = 0. That case should never happen though. This error happened one in every 3 or 4 runs previously when I was using a non-labeled version of the image to find the endpoints, but now that I am using the labeled image to find the endpoints, it is only happening once in a blue moon and I cannot figure out why.

[Ryan]

I have updated the question with the .mat data file. I did not add dbstop if error before this happened.

Answer 1

[Anton Semechko]

The error occurs for m=559. If you look at the object:

figure, imshow(K==559)
set(gca,'xlim',[481 580],'ylim',[500 600])

you will see that its a loop that does not have any end points. This is what ultimately causes the error.

[Ryan]

I assumed such when I did the unique/diff test in the original post. That implies either the multiplication is off or branchpoints is wrong in my opinion.

[Anton Semechko]

I don't know what your original post was, but I can tell you that multiplication has nothing to do with your problem. You should go back and reevaluate how you obtain 'V', because contrary to your assumption this image obviously contains isolated objects that simply do not have endpoints.

[Ryan]

I will upload the image to show you on Monday at the soonest. There is a chance I suppose for loops because the skeleton is coming from an inf- thin with bwmorph, but I do a thicken and re-thin to eliminate them.

[Anton Semechko]

That will be a different question ... I believe this one can be closed

Answer 2

[Image Analyst]

I don't know how this could happen. If rows is null, then endpts_labeled==m is completely false. And if that's true, then n must be zero and so the loop never gets entered. That is a contradiction. Try "dbstop if error" and then when it errors, see what n is. Alternatively you can try this (untested)

endpts_labeled = int32(zeros(size(K)));
endpts_labeled(endpts) = K(endpts);

instead of this:

endpts_labeled = immultiply(K,endpts);

in case the immultiply is giving you slight mismatches as referred to in the FAQ http://matlab.wikia.com/wiki/FAQ#Why_is_0.3_-_0.2_-_0.1_.28or_similar.29_not_equal_to_zero.3F

[Ryan]

Will do. I'll update this next time it catches it.

[Geoff]

Make sure you dump all your variables to a MAT file next time numel(rows) is zero =)

Answer 3

[Ryan]

I switched to .* over immultiply() and I haven't run into the error message again. This may not mean I am out of the woods yet, but for anyone else out there that may encounter a similar issue, I wanted the possible solution to be documented.

[Image Analyst]

immultiply may work if you cast it to int32 afterwards.

endpts_labeled = int32(immultiply(K,endpts));

just in case there are any really small fractional parts produced during the multiply because of quantization error.

[Steve Eddins]

K contains all integers, and endpts contains only 0s and 1s. There shouldn't be any quantization error going on here.

[Ryan]

Do you have an idea as to what would be causing the error then?