PROBABILISTIC DISTANCE APPROACH FOR SIMILARITY MEASURE IN SAR IMAGES

Abstract

Measuring similarity between two Synthetic aperture radar (SAR) images becomes challenging when there is no universal method used to measure the similarity in the two SAR images. Presence of speckle noise makes it difficult to follow the usual rule as for the optical images. Speckle noises are multiplicative in nature which makes it difficult to directly compare with another SAR images. For this reason it is interesting to study of several types of noises and their impact on the images. This thesis introduces the concept of Renyi entropy and Renyi mutual information in SAR images and its case study over different value of alpha and noise variance under the various noise models. Here we are trying to predict approximate similarity between two SAR images by calculating Renyi mutual information. We could observe that the measuring similarity is not an absolute measure as there are different sensors working over different climatic conditions and different temperatures. Here we tried to simulate the results with various noise models at different noise variance. It is very difficult to generalize a single method for all but the effort is to make it to the best possible results. The experiment to measure the similarity in SAR images is performed and validated.