Improved patch-based NLM PolSAR speckle filter based on iteratively re-weighted least squares method

The non-local means filters are popular and effective in speckle filtering as well as preservation of subtle details but have certain limitations. The limitation of NLM filter is its biased estimation in case of excessive speckle noise. Also, the similarity measure which preserves polarimetric or scattering property of data is the matter of interest in recent works. In this paper, a speckle filtering technique is presented which apart from reducing speckle noise, preserves polarimetric property, fine structures of the polarimetric SAR data and leads to an unbiased estimation in excessive noise. The filtering method adjusts weights iteratively, such that norm of distances is minimized. The proposed method is validated over two real PolSAR datasets (captured over San Fransisco Bay, USA & Mumbai coastal area, India by RADARSAT-2) and a synthesized SAR image. The qualitative and quantitative performance analysis is presented for the proposed method in terms of visual appearance of span, RGB PolSAR images, ENL and B-index. The proposed method is also evaluated for PSNR and SSIM on synthesized SAR image. It is found that the proposed method performs better than NLM filter and can be considered as a good alternative to NLM filters.