A Variational Framework for Adaptive Satellite Images Segmentation

Abstract

In this paper, we present an adaptive variational segmentation algorithm of spectral / texture regions in satellite images using level set. Satellite images contain both textured and non-textured regions, so for each region spectral and texture cues are integrated according to their discrimination power. Motivated by Fisher-Rao linear discriminant analysis, two region weights are defined to code respectively the relevance of spectral and texture cues. Therefore, regions with or without texture are processed in an unified framework. The obtained segmentation criterion is minimized via curves evolution within an explicit correspondence between the interiors of evolving curves and regions in the segmentation. The shape derivation principle is used to derive the system of coupled evolution equations in such a way that we consider the region weights and the statistical parameters variability. Experimental results on both natural and satellite images are shown.