A maximum likelihood framework for grouping and segmentation

Abstract

This paper presents an iterative maximum likelihood framework for perceptual grouping. We pose the problem of perceptual grouping as one of pairwise relational clustering. The method is quite generic and can be applied to a number of problems including region segmen­tation and line-linking. The task is to assign image tokens to clusters in which there is strong relational affinity between token pairs. The parameters of our model are the cluster memberships and the link weights between pairs of tokens. Commencing from a simple probability distribution for these parameters, we show how they may be estimated using an EM-like algorithm. The cluster memberships are estimated using an eigendecomposition method. Once the cluster memberships are to hand, then the updated link-weights are the expected values of their pairwise products. The new method is demonstrated on region segmentation and line-segment grouping problems where it is shown to outperform a non iterative eigenclustering method.