A Level Set Filter for Speckle Reduction in SAR Images

Despite much effort and significant progress in recent years, speckle removal for Synthetic Aperture Radar (SAR) image still is a challenging problem in image processing. Unlike the traditional noise filters, which are mainly based on local neighborhood statistical average or frequencies transform, in this paper, we propose a speckle reduction method based on the theory of level set, one form of curvature flow propagation. Firstly, based on partial differential equation, the Lee filter can be cast as a formulation of anisotropic diffusion function; furthermore, we continued to deduce it into a level set formulation. Level set flow into the method allows the front interface to propagate naturally with topological changes, where the speed is proportional to the curvature of the intensity contours in an image. Hence, small speckle will disappear quickly, while large scale interfaces will be slow to evolve. Secondly, for preserving finer detailed structures in images when smoothing the speckle, the evolution is switched between minimum or maximum curvature speed depending on the scale of speckle. The proposed method has been illustrated by experiments on simulation image and ERS-2 SAR images under different circumstances. Its advantages over the traditional speckle reduction filter approaches have also been demonstrated.