A new image segmentation technique using maximum spanning tree

Abstract

An alternative to the gradient-based image segmentation methods are those methods that use eigenvectors based on an affinity matrix built from pairwise pixel similarity. In this paper, we describe a new image segmentation algorithm using the maximum spanning tree. Our method works on the affinity matrix; however, instead of computing eigenvalues and eigenvectors, we show that image segmentation could be transformed into an optimization problem: finding the maximum spanning tree of the graph with image pixels as vertices and pairwise similarities as weights. The experimental results on synthetic and real data show good performance of this algorithm. © 2008 Springer-Verlag Berlin Heidelberg.