Contrast analysis based on homogeneous and non-homogeneous patterns for SAR image classification*

摘要

This study presents a novel contrast-based classification algorithm for Synthetic Aperture Radar image. The proposed algorithm primarily analyses homogeneous and non-homogeneous patterns within the moving window. The points which are either significantly brighter or darker than the central pixel are considered as non-homogeneous patterns. While the points, moderately equal to central pixel are described as homogeneous patterns. The mean intensity values of homogeneous and non-homogeneous regions are used thereafter for measuring the contrast around the central pixel of the moving window. The ISODATA algorithm is used to classify the contrast transformed image. The importance of this method is that it is computationally simple, works with minimum human interaction and robust to speckle noise. The validation of the method is done on RISAT (Radar Imaging Satellite) data.