Image segmentation by level set analysis

Abstract

This paper describes an automated image segmentation technique that subdivides regions of homogeneous texture. The method utilizes a level set analysis of scaled Gabor filter responses. Scaling is achieved via an area morphological process. Each scaled, filtered image is examined to locate important connected components based on minimal total internal variance and maximal edge localization. The candidate segments are selected using a granulometry of the gradient magnitude evaluated at the level lines of the connected components. The level set analysis avoids the high computational cost associated with conventional level set approaches by sampling only the significant level sets for processing. The target application for this segmentation technique is content based image retrieval.