Abstract

We present a new image segmentation algorithm based on graph cuts. Our main tool is separation of each pixel from a special point outside the image by a cut of a minimum cost. Such a cut creates a group of pixels around each pixel. We show that these groups are either disjoint or nested in each other, and so they give a natural segmentation of the image. In addition this property allows an efficient implementation of the algorithm because for most pixels the computation of is not performed on the whole graph. We inspect all ’s and discard those which are not interesting, for example if they are too small. This procedure automatically groups small components together or merges them into nearby large clusters. Effectively our segmentation is performed by extracting significant non-intersecting closed contours. We present interesting segmentation results on real and artificial images. 1

Citations

1824	Normalized cuts and image segmentation - Shi, Malik - 1997
1753	Dubes, Algorithms for Clustering Data - Jain, C - 1988
905	Fast approximate energy minimization via graph cuts - Boykov, Veksler, et al.
261	Segmentation using eigenvectors: A unifying view - Weiss
200	An optimal graph theoretic approach to data clustering: Theory and its application to image segmentation - Wu, Leahy - 1993
192	A maximum-flow formulation of the n-camera stereo correspondence problem - Roy, Cox - 1998
188	Combinatorial Optimization - Cook, Cunningham, et al. - 1998
126	WT: "A factorization approach to grouping - Perona, Freeman
76	Segmentation by grouping junctions - ISHIKAWA, GEIGER - 1998
21	Stochastic image segmentation by typical cuts - Gdalyahu, Weinshall, et al. - 1999