【摘要】 非局域均值(NLM)去噪算法假设图像噪声方差是定值,对于存在非线性噪声的图像处理效果不佳。像素滤波权重分配仅依赖像素块相似性,其准确性会受到噪声的干扰。为此,提出一种基于灰度-信息方差特性曲线的改进算法CNLM。该算法利用离散余弦变换,计算选取出特征像素点的有用信息方差及噪声方差,得到图像的灰度-纹理方差(IIVC)和灰度-噪声方差(INVC)2条特性曲线。2个像素点之间的权重受到这2点灰度对应的IIVC值的约束,以降低噪声对该过程的影响,算法中固定的噪声方差被每个像素灰度对应的INVC值替代以更好地去噪。实验结果证明,该算法对于加性和乘性噪声都有较好的噪声估计及去噪效果。更多还原

【Abstract】 Non Local Means(NLM) the assignment of the pixels’weight in smoothing parts depends too much on the patches’similarity,which is interfered by noise.The fixed variance parameter brings trouble in dealing with noisy image with nonlinear noise.This paper proposes an improved non local means(CNLM),with the help of intensity-variance characteristic curve.It adopts Discrete Cosine Transform(DCT) to analyze the texture variance and noise variance in image,and forms the Intensity-Information Variance Curve(IIVC) and Intensity-Noise Variance Curve(INVC).These two curves help CNLM to better calculate the weights between pixels and their variances.Experimental results show the effectiveness and efficiency of the proposed CNLM under various noise circumstances.更多还原