A new approach to SAR image speckle filtering

A new approach to Synthetic Aperture Radar (SAR) image speckle filtering combined with geometrical edge detection is presented. The traditional method detects edges and thin linear structures on windows with fixed size, and then filters the speckles in the sub-windows with fixed size. Speckle noise is not well suppressed along edges, lines or around point targets. The proposed algorithm first detects edges or lines on a series of local windows that have decreasing size, and then applies filtering to the statistics extracted from the corresponding sub-windows that do not cross the edges or lines. Thus the new algorithm is not only shape adaptive, but also window-size adaptive. In order to detect and preserve strong scatterers more efficiently, a decision threshold is chosen in this paper. Experiments show that the new algorithm is very efficient in speckle removal and in retaining edge sharpness and subtle texture details