Shape constrained image segmentation by parametric distributional clustering

Abstract

The automated segmentation of images into semantically meaningful parts requires shape information since low-level feature analysis alone often fails to reach this goal. We introduce a novel method of shape constrained image segmentation which is based on mixtures of feature distri-butions for color and texture as well as probabilistic shape knowledge. The combined approach is formulated in the framework of Bayesian statistics to account for the robust-ness requirement in image understanding. Experimental ev-idence shows that semantically meaningful segments are in-ferred, even when image data alone gives rise to ambiguous segmentations. 1.