The Classification Gradient
Vassili A. Kovalev, M. Petrou · 2006
We propose a method that uses bootstraping to classify the pixels in the two halves of a sliding window, assuming that if there is a real image boundary separating the two halves, the pixels in the two halves will be classified in two separate classes. The accuracy of the classification is used as a local "gradient". High values of this gradient allow us to detect weak statistical borders in 2D and 3D images