bogey4 wrote:
>
>
> Hi everyone,
>
> I'm conducting an experiment which aims to evaluate numerous
> binarization algorithms in Matlab. I've got about 25 algorithms up
> and
> running now, although I'm still trying to find a way to measure
> their
> results for comparison. I've got a "ground truth" image (drawn in
> photoshop) which I then degraded with noise and blurring, and then
> gave the degraded image to each binarization algorithm.
>
> Now I'm looking for a way of measuring how close each algorithm
> gets
> to producing the correct (ie ground truth) image. I've looked for
> m
> files that use the Hausdorff distance, although I can't seem to
> find
> any, so I was wondering if anyone could suggest a metric/distance
> or
> point me towards some source code that could be used to determine
> how
> closely the binarized image matches the ground truth image?
>
> Any help would be greatly appreciated,
>
>
A quick and nasty method might be to simply exclusive 'OR' the image
resulting from your test and the ground truth image. Areas where they
are different will contain active pixels, and those that are correct
will be inactive. Simply counting the number of active pixels will
give you some measure of the closeness of match. Thus if the two
images are identical, you will have a completely black exclusive OR
and will result in a count of 0, if the two are completely different,
you would have a count equal to the number of pixels in the images.
A quick way of counting the images is simply finding the mean
intensity of the exclusive OR image, then multiply this by the number
of pixels in the image.
This is probably not what you are looking for, but as no one else
responded, I thought I might as well ;)
