A Study Of SAR Image Speckle Suppression Based On Kernel Clustering And Optimal Iterative

Synthetic Aperture Radar(SAR) is a kind of radar which the small real antenna aperture can synthesis a larger equivalent antenna aperture with the relative motion of radar and target. SAR image has advantages of all-weather, all-day, high resolution, side-view imaging, and its characteristic signals are very rich which include many information such as amplitude, phase and polarization. In recent years SAR image processing(denoising, segmentation, object recognition, etc.) have got more and more attention. However, as a result of SAR imaging, it would produce coherent scattering echo so that SAR image contains random speckle noise, and the noise is multiplicative, which make different between SAR image denoising and general image denoising and affect the image quality and subsequent processing. So coherent speckle restrain of SAR image is necessary, and we should keep image details as possible.The main problem of SAR image denoising is to do good balance between removing speckle noise and retaining more image details. This article is aimed at making some improvements for speckle reduction methods in terms of retaining details of SAR image. Main work and contributions are as follows:1. Proposing a coherent speckle suppression method of SAR image which is based on kernel regression feature clustering and improved non-local means filtering. Due to the characteristic of the adaptive kernel regression function, it can represent some of the details features(edges, texture, etc.) of SAR image by weight value. In order to better deal with these characteristics, we adopt clustering processing, taking these features as the initial cluster centers, and then use the K-means clustering, then the similar feathers get together, so that we can get several clusters with similar characteristics, followed by improving the way of the similarity measure for non local mean filter, which effectively guarantees for each clustering. Non-local means denoising processing can be as much as possible to retain details of the SAR images.2. Proposing another despeckling method, which is based on diffusion and boosting adaptive iterative estimation for SAR speckle reduction method. Mainly to the introduction of a risk estimator based on minimum mean square error(MSE). For the diffusion and boosting iterative mechanisms, they have their own advantages and disadvantages. The later can make up for the shortcomings of the former, so we combine the two iterative methods, and use adaptive selection, to get the optimal number of iterations and the optimal type of iterative method. And then based on the results obtained in the optimal choice, we make non-local mean filter processing for SAR image. The method can achieve denoising purposeand get good retention of image details at the same time.