Suppression Of Speckle In SAR Images Based On Nonlocal Total Variation And Vision Computing

Synthetic aperture radar (SAR) is a all-weather, all-time, high-resolution, side-looking imaging radar, and has been widely used in many kinds of fields. However, its echo imaging mechanism unavoidably leads to speckle. In order to have a more efficient understanding and interpretation of SAR images, it’s necessary to suppress the speckle in SAR image before processing the SAR images. Local adaptive filtering can efficiently suppress the speckle while preserving the edges and details in the SAR images. These methods control the compromise of smoothing speckle and preserving the edges and details by the parameters, and have been one of the most commonly used methods. Most Local adaptive filtering based on the fully developed speckle assumption. However non-homogeneous areas in SAR image doesn’t satisfy this assumption. The parameter obtained by local information would make edges and textures vague, even lead to deterioration of the local image quality. Enhanced Lee filter gets a division of SAR image by the local statistical information, and then adopts different speckle suppression methods in different regions. But the division in this way does not take fully into account the global structure information of image, which lead to the phenomenon that uneven area appear in a large homogeneous area.This paper gets a division of SAR image by primal sketch sparse representation model in Vision Computing, and adopts different speckle suppression methods in different regions. The main work of this paper is as follows:(1) The primal sketch in the primal sketch sparse representation model uses dots and segments to approximate points, edges and ridges of SAR image, which correspond to the non-homogeneous area of the image, including the image structure and direction information. But the model using one pixel wide segments to represent the line tartgets’ locations and diretcions would generate deviation. So geometry structure blocks are put forward to get region mark, and then to divide the SAR image into homogeneous regions, non-homogeneous areas according to the region mark.(2) Lee filter of a adaptive window is adopted to suppress speckle in homogeneous regions. The largest window is token for image filtering. Gradually resize the window until the window does not contain the structure of the image, which is the right size to filter by Lee filtering in the same window. And we assume that the non-homogeneous areas satisfy the additive noise model, and adopt a method of suppressing speckle based on nonlocal total variation and vision computing, in which the ROF model of total variation can well describe the edges of images, combinated nonlocal mind. Simulation results demonstrate our method achieve a balance between speckle suppression and edge preserving.