Question

I am working on image segmentation and i want to evaluate my work using misclassification error which equals

ME = 1- (|b0 intersect bt|+ |f0 intersect ft|)/(|b0|+|f0|)

where BO and FO denote the background and foreground of the original (ground-truth) image, BT and FT denote the background and foreground area pixels in the test image.

i don’t know how to discriminate object from background, and how to generate a scaler value.
I am working on matlab. I will appreciate any help. it is urgent.

Answer 1

Using the nnz command, you can easily count the number of non-zero pixels in an image. Assume you have a segmented image si and a ground truth image gt, both of which are binary images, with ones wherever there is signal, the mis-classification can be written as

ME = 1 - (nnz(~gt & ~si) + nnz(gt & si))/(numel(gt));

In the denominator, you just put the total number of pixels, since that’s the sum of background plus foreground.