Graph Spectral Segmentation of SAR Image Based on Information Similarity Measure

In this paper, an information theory based similarity measure, calculable from synthetic aperture radar (SAR) image based on pixels, but relying on global image statistics, is proposed. It considers the texture of speckle noise which often carries useful information about the scene being imaged, thus can capture subtle variations of scene texture whenever they are embedded in a heavy speckle. Graph spectral segmentation method based on the information similarity measure is carried out on two real SAR images. Experimental results are very encouraging compared with the commonly used similarity measure.