A new technique for color based image segmentation using support vector machines

Abstract

Due to the large requirement in spacial imagery generating and updating geographical information, the segment-based image analysis methods are becoming more popular. Hence, this paper presents a new algorithm for image segmentation (for LANDSAT images) using low-level feature, i.e., color using Support Vector Machines. In the first step using de-correlation stretching color information is separated out from the satellite image and in the second step using M-estimator based support vector machine clustering algorithm the regions are grouped into a set of classes. A new M-estimator is proposed for robustification. Simulation results are provided to demonstrate the efficacy of the proposed support vector machine based clustering algorithm for color based image segmentation.