Image Segmentation based on Normalized Cut from the Perspective of the Discriminant Information

Image segmentation is a fundamental problem in computer vision. Normalized Cut (Ncut) scheme uses second smallest eigenvector of a special matrix to solve this problem. In this paper, firstly, it is shown that optimization of Ncut (as an unsupervised method) is equivalent to optimization of Fisher-Rao criterion (as a supervised method) in classification. Then, the classification experience is used to gain a new perspective on the order and selection of eigenvectors in Ncut approach. Experimental results on image segmentation, demonstrate the truth about this alternative view of eigenvector selection which leads to less amount of Ncut for image segmentation.