bug in source code histogram.cpp (Feature #3108)

According to the formula here: http://docs.opencv.org/modules/imgproc/doc/histograms.html?highlight=hist#cv.CompareHist, I belive that the implementation is correct:

    if( method == CV_COMP_CHISQR )
    {
        for( node1 = cvInitSparseMatIterator( mat1, &iterator );
             node1 != 0; node1 = cvGetNextSparseNode( &iterator ))
        {
            double v1 = *(float*)CV_NODE_VAL(mat1,node1);
            uchar* node2_data = cvPtrND( mat2, CV_NODE_IDX(mat1,node1), 0, 0, &node1->hashval );
            double v2 = node2_data ? *(float*)node2_data : 0.f;
            double a = v1 - v2;
            double b = v1;
            if( fabs(b) > DBL_EPSILON )
                result += a*a/b;
        }
    }

Li, please explain why do you think that the code is buggy, otherwise this ticket will be cancelled.

This issue keeps popping up, see #1263, #2199. There's definitely two variants to Chi-square metric, depending on its intended purpose.

I started working on a patch where both versions coexist. It's clean and working but I failed to fix the accuracy/regression tests, so the pull request 484 (https://github.com/Itseez/opencv/pull/484) was closed.

It'd be great if you could pick up the work and create another pull request for master branch (it must be there because it changes API). Then it will be part of next big OpenCV release. At the moment, I really don't know when I get back to looking at the patch again. With a fresh look at it though, it might be fixed quickly.

Thank you Li for submitting the ticket and thank you Sebastian for your Pull Request, we really appreciate it!

But unfortunately we have highly limited human resources at the moment and if there is a chance you can find an extra time (I hope it will take less time for you then for anybody else, since you have made almost all the work) to finish this PR that would be great!

Thank you in advance,
Victor.