一种利用局部块相似均值去噪的方法

摘要

Non Local Means(NLM) is a state-of-the-art method for image denoising based on a nonlocal weighted mean of image blocks' similarity.It performs well on Gaussian noisy images.However, local extremal points are prone to yield in smoothing area.By analyzing strengths and weaknesses of NLM's searching area and similarity function,the reasons of extre mal points' appearance in smoothing area are given.A novel image denoising method based on NLM framework is presented to achieve improved performance.The new method adopts a compact weight set by using a threshold based on image block variation to eliminate irrelevant similar blocks.Compared with the original NLM,the new method is more efficient in keeping smooth in homogeneous areas and boundary maintenance according to the experimental results.%非局部平均(NLM)是一种基于图像块之间相似性的加权平均去噪算法,对高斯噪声具有很好的抑制作用,但是在平滑区域的去噪效果并不是很好.从相似块的搜索区域和相似性度量函数两个方面对NLM算法进行了分析,指出其在平滑区域容易产生极值点的原因.提出了一种结合图像块特征的阈值方法,用于消除搜索区域中的无关图像块,提高了图像相似结构的利用率.实验表明,新算法对光滑区域和细微结构的去噪能力要优于NLM算法.