SAR Speckle Denoising Based on Statistic Model Combined with Medication to Significant Wavelet Significant Coefficient
Yu Qiu-ze①②; Zhu Guang-xi①; Liu Jian②; Tian Jin-wen②; Mao Hai-cen②
①Department of EE, Huazhong University of Science and Technology, Wuhan 430074, China; ②Key Laboratory of Education Ministry for Image Processing and Intelligence Control, Huazhong University of Science and Technology, Wuhan 430074, China
Abstract:This paper proposes a new method based on statistical model of wavelet coefficients combined with modification to them according to significant coefficient rule. In the method, wavelet coefficients of logarithmic image are firstly modeled as mixture density of two Gaussian distributions with zero mean. In order to incorporate the spatial dependencies into the denoising procedure,Hidden Markov Tree (HMT) model is explored and Expectation Maximization (EM) algorithm is proposed to estimate model parameters. Bayes Minimum Mean Square Error (Bayes MMSE)method is used to estimate the wavelet coefficients free of noise. The wavelet coefficients are updated according to a rule whether the coefficient is a significant one or not. 2D inverse DWT and exponential transform are performed on the updated coefficients to get denoised SAR image. Experimental Results using real SAR images demonstrate that the method can not only reduce the speckle but also preserve edges and radiometric scatter points.