Computational techniques in the visual segmentation of static scenes

Abstract

A wide range of segmentation techniques continues to evolve in the literature on scene analysis. Many of these approaches have been constrained to limited applications or goals. This survey analyzes the complexities encountered in applying these techniques to color images of natural scenes involving complex textured objects. It also explores new ways of using the techniques to overcome some of the problems which are described. An outline of considerations in the development of a general image segmentation system which can provide input to a semantic interpretation process is distributed throughout the paper. In particular, the problems of feature selection and extraction in images with textural variations are discussed. The approaches to segmentation are divided into two broad categories, boundary formation and region formation. The tools for extraction of boundaries involve spatial differentiation, nonmaxima suppression, relaxation processes, and grouping of local edges into segments. Approaches to region formation include region growing under local spatial guidance, histograms for analysis of global feature activity, and finally an integration of the strengths of each by a spatial analysis of feature activity. A brief discussion of attempts by others to integrate the segmentation and interpretation phrases is also provided. The discussion is supported by a variety of experimental results.