The Research On Non-locak Means Flitering Algorithm

Generally, in the process of image acquisition and transmission, due to the limitationsof the equipment used and the transmission channels, noise is often introduced which affectsimage’s visual effect and even hinders people’s normal recognition. Therefore, in order toimprove the quality of images and build the foundation for following process, imagedenoising is the indispensable part in image processing.Image Denoising based on Adaptive BM3D and Singular Value Decomposition(SVD-ABM3D) is studied in this paper. The main research work includes:1. An improved version of BM3D with adaptive threshold is proposed. As BM3Dperforms the fixed hard-thresholding operator when comparing the similarity between twoblocks, it couldn’t pick the best matching threshold which leads to the fall of denoisingperformance. The proposed adaptive matching threshold alogorithm adjusts itsblock-matching threshold according to the different images. This paper makes use ofStructure Similarity index (SSIM) to find the most suitable matching threshold and thespecific steps is as follows: firstly the standard image is divied into blocks, and pointwisecalculation of SSIM and Eulidean distance of BM3D is performed. Then the most suitablematching threshold with maximun SSIM value is obtained. The same block division to thenoisy image and the calculation of every block’s nosie level and gradient value are adopted.Finally, data fitting to the obtained data to build the adaptive threshold fomula is applied.Experiment results are given to show that the proposed algorithm achieves better denoisingperformance than the original BM3D with sharp edges and smooth flat region.2. Based on adaptive threshold BM3D, this paper proposes the adaptive BM3Dalgorithmn with singular value decomposition pre-flitering (SVD-ABM3D). Beforeapplying A-BM3D, noise level of noisy image is estimated. To the strong noise image,pre-filtering based on SVD is performed. Then SVD reconstruction with suitablereconstruction number is applied. Finally, A-BM3D to denoise the reconstruction image isadopted. As SVD based pre-flitering is lossy, so applying SVD based per-flitering to weaknoisy image may lead to bad result. Experiment results show that SVD based pre-filtering isfeasible and effective as well and the proposed algorithmn further improves the performanceof A-BM3D.