Denoising Synthetic Aperture Radar Image Via Wavelet Transform

Synthetic aperture radar (SAR) imaging system was invented in the 50s and 60s of 20th century. One of the most important features of SAR imaging system lies in that it can produce high resolution images of large area under all-weather condition. Therefore, it is widely applied to numerous fields such as high quality mapping, earth surface surveillance, search and rescue, mineral resources detection, automatic target recognition, and etc.One of the main problems with SAR imagery is that the images generated are inevitably disturbed by multiplicative speckle noise. This is due to coherence of backscattered electromagnetic waves which produce the image data. Speckle noise in SAR image blankets image details and deteriorates image qualities, and may seriously impair the effects of automatic scene segmentation and target recognition. Since the cause of speckles is very special, general denoising approaches employed in digital image processing are not very applicable. Thus, reducing and suppressing speckles have always been a major concern of the SAR image community.The requirements of denoising SAR image include not only filtering out noise but also preserving image details like edges and textures. Research scientists have already proposed several spatially adaptive filters which take advantage of the properties and features of speckle noise. Whereas, this thesis adopts a multiresolution point of view, applying wavelet transform, an extremely powerful mathematic tool, to denoising SAR image.In this thesis, first there is a very brief introduction to synthetic aperture radar imaging principles. Then, the cause and statistical model of speckles in SAR image are analyzed. After that, a generalized description of wavelet transform and multiresolution analysis is presented. Based on these fundamentals, the probability distribution of wavelet coefficients of SAR image is investigated and then the classic spatially adaptive wavelet soft-thresholding scheme used for denoising natural image is employed to suppressing speckles. Furthermore, we propose a new approach to calculating the threshold and estimating related parameters according to local statistics of wavelet coefficients. Experiment results on actual SAR images show that the proposed method makes a good tradeoff between noise suppression and fine detail preservation. The denoised testing SAR images are superior to those processed by traditional spatial filters when they are subjected to both visual perception and numerical evaluation. At last, there is a short but insightful discussion of the significance of wavelet transform and multiresolution analysis to SAR image denoising.