Refined adaptive speckle filtering for SAR images

The presence of speckle in SAR images reduces the ability to perceive fine details and rules out a visual perception of systematic structures that would make it possible to describe a specific feature of the earth's surface and makes the radiometric and textural aspects less efficient for class discrimination . Many adaptive filters have been developed for speckle reduction and they are based on a test related to the local coefficient of variation of the observed image, which describes the heterogeneity of the scene. In this paper, the well known non linear adaptive filters such as Frost, Wu, and Nezry filters are presented, and considering the limitations of the analysed filters and the good results achieved using the ratio edge extraction, a modified adaptive speckle filter has been implemented based on both geometrical and statistical properties (coefficient of variation) of the local area. The procedure has depended on several parameters such as window size, the width of the linear features, the thresholds, the values of Cu (coefficient of variation of multiplicative noise) and C-max(the maximum value of the distribution of C-upsilon). The successful use of the ratio edge and line detectors has been employed in adaptive speckle filter, in order to apply the enhanced frost version in every window or subwindow selected. As was expected, the new filter performs better, i.e., we can achieve very good smoothing in the homogeneous areas preserving, at the same time, structural and textural information, edges, linear features. The filters are tested on a ERS-1 SAR image corresponding to an area of southern Algeria called Laghouat. The good results obtained and the improvements with respect to die classic algorithms are presented and discussed.