Multiregion competition: A level set extension of region competition to multiple region image partitioning

Abstract

The purpose of this study is to investigate a new representation of a partition of an image domain into a fixed but arbitrary number of regions by explicit correspondence between the regions of segmentation and the regions defined by simple closed planar curves and their intersections, and the use of this representation in the context of region competition to provide a level set multiregion competition algorithm. This formulation leads to a system of coupled curve evolution equations which is easily amenable to a level set implementation and the computed solution is one that minimizes the stated functional. An unambiguous segmentation is garanteed because at all time during curve evolution the evolving regions form a partition of the image domain. We present the multiregion competition algorithm for intensity-based image segmentation and we subsequently extend it to motion/disparity. Finally, we consider an extension of the algorithm to account for images with aberrations such as occlusions. The formulation, the ensuing algorithm, and its implementation have been validated in several experiments on gray level, color, and motion segmentation.