Comparison of speckle noise reduction techniques using near shift invariant wavelet transforms with application to SAR imagery

Synthetic aperture radar images are affected by multiplicative speckle noise that make it difficult for automatic classification and visual analysis of targets. This paper presents a framework for contrasting various wavelet shrinkage schemes found in literature using three recently developed classes of wavelets, for the despeckling of SAR images. Each of the wavelet transforms is near shift invariant and forms an over-complete basis. The properties and design procedure of these wavelets are discussed, revealing their advantages over the standard discrete wavelet transform. Image de-noising experiments are conducted on synthetically simulated speckle and real SAR images and recommendations are made on choosing the appropriate shrinkage scheme for a particular application.