Study On Denoising Algorithm For SAR Images Based On Wavelet And Bilateral Filter

Synthetic Aperture Radar (SAR) is a kind of radar system, which can obtain high resolution image and is widely used in various fields such as agriculture, forestry, and military. The SAR images are polluted by noise inevitably in the imaging process due to the SAR images unique imaging principle, which brings serious influence in the subsequent target detection and identification.In this thesis, the related method of SAR image denoising and the basic theory of wavelet transform and bilateral filter are studied firstly. Then aiming at the problems in the existing denoising method, which not have the character of denoising and edge-preserving and the denoised image generate the pseudo Gibbs phenomenon. We combine the algorithm of Wavelet transform and bilateral filter, and intensive study the related improved algorithms. The main work can be summarized as follows:(1) A SAR image denoising algorithm based on wavelet threshold and bilateral filtering is proposed. Due to the traditional wavelet denoising do not have the character of denoising and edge-preserving. The algorithm firstly improves the traditional wavelet threshold estimation method, the new threshold has better adaptability. Then let the processed wavelet coefficients reconstruct to the first layer, and the reconstructed high-frequency images use Birge-Massart strategy to obtain threshold and are denoised again. The low frequency image is denoised using bilateral filtering. Experiments show that this algorithm can inhibit SAR image speckle noise and keep the details of the image better.(2) A SAR image denoising algorithm based on double domain image denoising (DDID) and Cycle Spinning is proposed. For the DDID algorithm can make the denoised image generate the pseudo Gibbs phenomenon, leading to the man-made texture or local distortion. We use Cycle Spinning algorithm to shift the image firstly, then using the DDID algorithm to denoise the shifted image signal and anti-shift the denoised image signal to remove pseudo Gibbs phenomenon. Compared with other methods, the denoised SAR image of the proposed algorithm is smoother and has fewer artificial textures.(3) A SAR image denoising algorithm based on based on double domain image denoising (DDID) and NL-Bayes is proposed. For DDID algorithm needs three iterative computation, making the algorithm performance is poor. So we use NL-Bayes approach to denoise the guiding image of DDID algorithm, then perform an iteration computation to get the final denoising results. Experiments show that the algorithm has higher execution efficiency and can solve the pseudo Gibbs phenomenon of DDID algorithm, the denoised image is smoother and more natural.