中文摘要：

针对非局部均值（NLM）去噪算法运算量大的缺点,给出一种快速NLM方法。通过特征来表示块的信息,使每个块只计算一次,从而减少计算时间。与先前的方法相比,所给方法的相似性度量准则与NLM方法一致,且不需要预先设置阈值。为加速相似块的搜索,采用2D直方图和积分图两种方法。另外,给出基于块的方法对图像噪声进行盲估计,以求解滤波参数。针对坦佩雷图像数据库（TID）所进行的实验结果表明,所给方法利用整幅图像信息,故其性能明显优于半局部均值（SLM）算法。

英文摘要：

Large computation is an issue for Non-Local Means（NLM）method.A fast method is proposed to implement the NLM method by representing patches in terms of features so that each patch is computed only one time.Unlike the previous methods,the metric for patch similarity is the same measure as used in NLM instead of heuristic thresholds.2Dhistogram and summed-area table are employed to speed up searching similar patches.Moreover,apatches-based method is developed to blindly estimate noise level for filtering parameter.The proposed method achieves NLM solution in entire image.Experiments carried on Tampere Image Database（TID）demonstrate that the proposed method outperforms the semi-local implementation,which restricts patch search in a local neighborhood.