LOCAL HISTOGRAM BASED CLASSIFICATION OF SAR IMAGES USING SPATIALLY DEPENDENT MIXTURES

Abstract

We propose a local histogram based mixture model for synthetic aperture radar image classification. In this model, local histograms are modeled by multinomial densities. The model has also a spatially dependent class label prior to achieve a smooth contextual classification. Based on the classification results obtained on real TerraSAR-X image, it is shown that the proposed model is capable of more accurately classifying pixels especially in the heterogeneous regions, like urban areas, compared to conventional mixture model.