The Research On Speckle Reduction For Synthetic Aperture Radar Images

Synthetic aperture radar(SAR) emits the coherent electromagnetic waves, and receives the reflected echoes from the target surface to image. It has the features of high resolution, all-day, all-weather, strong penetrability, covering a large area and so on, so it has been widely used in both military and civilian areas. Because of the imaging theory, SAR images have the inherent speckle noise. Speckle noise seriously hampers the further interpretation and application of SAR images. To overcome the challenges of speckle reduction of SAR images, such as parameters configuration in bilateral filtering, edge detection in methods based on anisotropic diffusion and edge preserving and speckle reduction level in methods based on total variation, this dissertation carries out the research on the challenges, and then obtain the contributions follows:To overcome the challenge of parameters configuration of the spatial closeness variance and gray value similarity variance in bilateral filtering, this dissertation proposes the spiral-iterative method. Given the initial value of gray value similarity variance, the method spiral iteratively computes the optimal gray value similarity variance based on the characters of the curves of speckle reduction evaluation indexes. Compared with the traditional method, experimental results show that the accuracy of variance of gray value similarity is increased by one order of magnitude in the case of almost the same computing time; the computing time is reduced by one order of magnitude for the same accuracy of variance of gray value similarity. To overcome the initial dependence of spiral-iterative method, the cross-iteration method is proposed. Given two initial values of gray value similarity variance, the method iteratively constructs two straight lines, the intersections of the lines converge to the optimal the value of gray value similarity variance, and the convergence of the method is demonstrated. The method not only solves the initial dependency problem, but also has faster convergence, and gray value similarity variance has higher accuracy.To overcome the challenge of edge detection in methods based on anisotropic diffusion, this dissertation proposes the method of speckle reduction based on the image entropy anisotropic diffusion. The method improves the accuracy of edge detection based on the uniformity of the energy distribution in space. The method does not require the evaluation of the variance and the mean in the traditional methods, and avoid the influence to edge detection due to the estimation error of the statistics. So it obtains the better effect of speckle reduction.To overcome the challenge of edge preserving and speckle reduction level in methods based on total variation, this dissertation proposes a new model for speckle reduction, and give the numerical algorithm to solve the model. The model integrates non-convex regularizer and spatially adaptive regularization parameters, can better maintain the edge and texture details of image, and reasonable controls speckle reduction level of the different regions. Compared with the model only using non-convex regularizer or only spatially adaptive regularization parameters, the model can get better effect of speckle reduction.