An improved non-local means de-noising approach

Three modifications are proposed for the Non-Local Means (NLM) de-noising algorithm. The symmetric property existing in many images is exploited first, symmetric transformations are introduced in neighborhood comparison so that image self-similarity property can be made better use of; Then an adaptive filtering parameter selection method is proposed based on image intensity statistics; Besides, a two-stage Non-Local Means filtering method is proposed adopting the structure preserving property of NLM algorithm. De-noising results for test images demonstrate that compared with the original algorithm, the modified one can remove noise more efficiently with image structures well preserved, the PSNR can be improved by 5. 9dB at most, and the de-noising performance outperforms that of the BM-3D algorithm.