A ROBUST CLUSTERING ALGORITHM FOR SAR IMAGE SEGMENTATION USING HYBRID SPATIAL INFORMATION

Abstract

Synthetic aperture RADAR (SAR) is one of the primary sensor used to perform the task of remote sensing for various applications. All weather and day and night operation of SAR makes them advantageous to their optical counterparts. SAR has been used for many applications like DEM generation, deforestation, activity monitoring, military surveillance, volcano eruption monitoring etc. High-resolution images of large areas are obtained by synthetic aperture radar imaging techniques. Spatial orientation, dielectric constant and roughness of the imaging area affects the intensities of pixels in a SAR image. Being an active sensor, SAR depends on its own transmitted energy. It generates the images of area based on the return scattered by it back to the radar’s antenna. Raw SAR signal data is processed to generate spatial image. The SAR imaging can be performed from satellite or airborne platform with a side looking antenna. SAR images are inherently contaminated by speckle noise, which is multiplicative in nature. Speckle is caused by interference of coherent wave fronts .Being an active imaging system, SAR images suffer from the inherent multiplicative noise known as speckle, which originates from the interference of the coherent wave fronts. The presence of speckle makes image processing tasks challenging for SAR images [1]. Segmentation of image is one of the most critical pre-processing step done before classification and identification of different regions and objects present in the image. Segmentation of image is to partition it into regions (also called segments, classes or subsets) which are similar with respect to one or more characteristics or features e.g. grey tone or texture. The performance of conventional intensity based image segmentation techniques deteriorates for speckle contaminated SAR images. Various speckle reduction techniques like spatial filtering and multi-look processing can be used, but it will reduce the resolution of image. Requirement of high resolution in remote sensing application make performing segmentation in presence of speckle inevitable. Given the challenge of segmenting SAR images in presence of speckle, thirteen image segmentation techniques are explored in this study. For speckle corrupted SAR images, intensity is not a suitable image feature for segmentation purposes, thus seventeen 2 different texture based image features have also been explored in the study. As a results, a total of 221 feature-segmentation technique combinations have been analysed in the study.