Research On Synthetic Aperture Radar Image Despeckling Algorithm

Synthetic Aperture Radar(SAR) images have been widely used in the fields of national economy and national defense science and technology, and speckle suppression is a very important pre-processing for SAR image. In order to better interpret SAR image, a good despeckling algorithm should not only suppress speckle noise effectively, but also preserve the texture and edge details.This paper studies the theory and technology of SAR image speckle reduction. Firstly, SAR imaging principle and characteristics of speckle noise are introduced briefly, the development of SAR image despeckling is reviewed, and the quality evaluation index of SAR images is discussed. The main works of this paper are as follows.(1) A wavelet domain SAR images speckle reduction algorithm is proposed based on generalized likelihood rate. Bayes shrinkage function expression is derived under the framework of joint detection and estimation. More specifically, redundant wavelet transform is performed to speckle SAR image to obtain the mask of each wavelet coefficient. We use scale exponential distribution and Gamma distribution, respectively, to model the likelihood conditional probability of speckle noise and useful signal. According to the mask, the parameters of the two modes are estimated by maximum likelihood estimation method, and thus the likelihood conditional probability ratio is calculated. Experiment results show that the proposed method can effectively filter the speckle noise, and at the same time preserve the image details as possible. Satisfactory results are achieved on both synthetically speckled images and real SAR images.(2) Over smoothing of image details is common in despeckling procedure for real SAR image. This paper presents a heterogeneity based despeckling algorithm to address this issue. Homogeneous region and important texture detail region can be precisely distinguished through heterogeneity measurement. The wavelet coefficients are classified, and coefficients in each class are proportionally adjusted accordingly. After proportional adjustment, generalized likelihood rate based Bayesian speckle reduction method is used to improved despeckling performance. In the end, the research of the paper and the shortcoming are summarized, and the future directions is also pointed out.