Application and Analysis of Advanced Denoising Algorithms on SAR Data

Abstract

Remote Sensing is the science that made it possible to acquire information about objects or events from a distance. Satellites and aircrafts are generally used for this purpose. Radio Detection and Ranging (RADAR) is the method used to acquire such information through the received signal. One such type of imaging RADAR that is commonly used is the Synthetic Aperture Radar (SAR). SAR helps to create 2-Dimensional or 3-Dimensional images of the objects such as landscapes. But it often gets affected by a peculiar type of noise called speckle that appears to be granular in nature. Speckle noise severely affects the spatial statistics of the image and makes it difficult to interpret the image accurately. Thus denoising algorithms such as Gamma MAP, Lee, Frost, Kuan, Sigma and their modified variants are called standard speckle filters. They are needed to minimise speckle noise. In some of the filtering algorithms like Gamma MAP, Enhanced Lee and Enhanced Frost, the image is first divided into 3 regions - homogeneous where the spatial statistics of the pixels are similar, heterogeneous where the spatial statistics vary largely and large scattering regions where the pixel values are high, and then apply the filtering algorithm based on this classification. These filters are based on coefficient of variation. Further for the study and thorough understanding of the advanced denoising algorithms the study of standard speckle filters is very much needed. Block Matching 3 Dimensional (BM3D), Refined Gamma MAP, Speckle Reducing Anisotropic Diffusion (SRAD) and Wavelet Based Thresholding filters are some of the advanced speckle filters. The above mentioned filters have been implemented in MATLAB 2013a software as well as C language. The above filters are also compared according to different quality assessment parameters. It is important to analyze and compare the end results of the filters. An ideal filtering algorithm would not only reduce speckle effectively, but also preserve the mean value, edges, lines, features and at the same time result in high Signal to Noise Ratio. Thus it is important for the complete evaluation of the filtering algorithm. The C source codes of the filters have been incorporated in an ISRO developed indigenous software called Microwave Data Analysis Software (MIDAS) which is a multi purpose software. Apart from filtering, it is also used for classification of glaciers, agricultural fields, conversion of data from one form to another form and generating scattering matrices. These are some of the few applications of MIDAS to be named.