Research On SAR Image Speckle Reduction Based On Non-Local Means

Synthetic aperture radar(SAR) is a coherent imaging radar system, and the product of the coherent imaging SAR image has random speckle which has the characteristic of multiplicative noise. The existence of speckle greatly decreased the SAR image processing performance of interpretive system, so the speckle reduction of SAR image becomes one of the most important researching fields of SAR image processing. At present, many approaches of filtering algorithm are proposed in spatial domain, transform domain, and partial differential anisotropic diffusion, but most of them are based on local mean techniques. Hence, a non-local means(NLM) algorithm is used in the speckle reduction research of SAR images in this paper.In this paper, five aspects of research is proceeded around speckle reduction of SAR image based on NLM techniques:First, a fast realization method of classical NLM algorithm is presented by analyzing the characteristics of traditional NLM algorithm. With the cost of memory space, the noise elimination is maintained and the real-time effect of algorithm is more effective.Second, two NLM speckle reduction algorithms of SAR image which based on the measure of spatial similarity are presented by the introduction of traditional NLM algorithm to classical speckle reduction algorithm of SAR image in spatial domain. The first one based on variation coefficient uses the traditional NLM technique to estimate variation coefficient ratio and mean ratio between the current pixel and the similar window of other pixels in the neighborhood, and replace the Gaussian weighted Euclidean distance with the product of the two alternatives. Then, improving this algorithm with the attenuation factor of Frost filter. At last, the negative exponential form is converted into weight, the weighted average can implement the speckle suppression of SAR image. The second algorithm which based on Kuan filter coefficient use the Kuan filter to pre-process SAR image, and obtain the Kuan filter coefficient by calculating the information entropy, then use the traditional NLM technique to estimate the mean ratio between current pixel and the similar window of other pixels in the neighborhood and replace the Gaussian weighted Euclidean distance with the product of two alternatives. Finally, a new similarity measure parameter is established by combining the attenuation factor of Frost filter and negative exponential form in which it is converted into weight, and the weighted average can implement the speckle suppression of SAR image. The two methods have better performance on the homogeneous area for speckle reduction, and the smoothing effect of second algorithm is better than the first one. Furthermore, the information protecting ability of the edge region is also improved.Third, combining traditional NLM and the speckle reduction algorithm of SAR image in wavelet domain,the a new NLM speckle reduction algorithm which based on wavelet domain similarity measurement is proposed. Firstly, this algorithm obtains the low frequency information of SAR image which used the wavelet decomposition, then calculates the entropy of information. Secondly, traditional NLM technique is used to estimate mean ratio between the current pixel and the similar window of other pixels in the neighborhood, the Gaussian weighted Euclidean distance is replace by the product of two alternatives. Finally, a new similarity measure parameter is established by combining the attenuation factor of Frost filter and the negative exponential form which is converted into weight, and the weighted average can implement the speckle suppression of SAR image. From the experiment, we can see that this approach has a better smoothing performance on the homogeneous area, and has a larger breakthrough on the edge detail maintaining aspect.Fourth, by the dint of NLM which based on similarity measurement of the variable metric window and wavelet domain, a NLM algorithm for SAR images which based on the variable metric window is presented. According to the algorithm, the input image is separated by a threshold of the noised image which based on the ESM of Gaussian gamma window. Then the homogeneous region and the edge region adopt different scales respectively with the consideration of the analysis of the search window size and the similar window on the filtering effect. Last, speckle reduction is accomplished by NLM which based on the similarity measurement of variable metric window and wavelet domain. The algorithm has a better performance of smoothing in homogeneous area, and the information of edge detail is protected too. It is greatly demonstrated that, from the noise reduction experiment of the real SAR image, the performance of edge preservation is better than other algorithms.Fifth, a GUI software of SAR image speckle reduction algorithm is realized by mixing the MATLAB program and Lab VIEW program. The software used several methods of this paper to reduce the speckle of SAR image, which can not only obtain the de-noised image and edge image, but also the performance evaluation parameters after noise reduction.In this paper, the research background and image-forming principle of SAR image is introduced at first, and then the traditional NLM and its fast realization way is proposed. After that, two algorithms in spatial filtering aspect and one algorithm, one improved algorithm in wavelet domain are presented. At last, the GUI software of SAR image speckle reduction algorithm is realized by Lab VIEW, then a summary and expectation of the full contents are conducted.