Discontinuity Adaptive Non-Local Means With Importance Sampling Unscented Kalman Filter for De-Speckling SAR Images

This paper brings out two extensions/utilities of well-known non-local means filter. Firstly, we propose an improvement to basic non-local means filter (NLMF) by incorporating discontinuity adaptive weights, instead of Gaussian weights for similarity assessment of patches in NLMF. As against the NLMF, discontinuity-adaptive NLMF (DA-NLMF) exhibit better edge and detail preserving capability, while providing very effective de-noising at the homogeneous areas. Secondly, we propose to apply DA-NLMF on recently proposed Importance Sampling Unscented Kalman Filter (ISUKF) [1], which can also be thought of as an elegant post processing technique, in general. Specifically, we propose an efficient method for de-speckling Synthetic Aperture Radar imagery by combining ISUKF and DA-NLMF. In our approach, the DA-NLMF provides an efficient de-speckling as well as feature preservation, when its parameters of its (DA) weighting function are derived from ISUKF results. The performance of these methods is demonstrated on both synthetic and real examples, and the proposed method gives excellent results than the standard speckle filtering methods as well as various advanced methods.