Gaussian weights Euclidean distance between pixel blocks with a high degree of similarity in the neighborhood is calculated to estimate current pixel value in fast non-local means algorithm,which achieves good results in the low-frequency part of the image,but loses partial edge information in the high-frequency part of the image because similarity between pixel blocks cannot be reflected by Gaussian weights Euclidean distance effectively. In order to retain more information in the high-frequency part of the image, a new weighting function is constructed,in which feature similarity color index composed of phase congruency,gradient and chrominance infor-mation is introduced into Gaussian weights Euclidean distance of fast non-local means algorithm and a denoising method is proposed based on it. Similarity between pixel blocks is computed by this new weighting function. Thus the estimated points are acquired with all pixels in three channels of color image filtering with block-by-block and averaged to obtain the entire filtered image. The experimental results show that compared with fast non-local means algorithm,it has improved the PSNR,FSIMC and retained more detail.%快速非局部均值算法利用像素邻域内具有高相似度的像素块之间的高斯加权欧氏距离来估计当前像素值,对于低频图像部分可取得较好的效果,而对于图像的高频部分,因计算获得的高斯加权欧氏距离不能很好地反映图像块间的相似性,会导致图像部分边缘信息的丢失.为了保留图像中更多的高频信息,在快速非局部均值算法中高斯加权欧氏距离的基础上,引入由相位一致性、梯度及色度特征组成的彩色图像特征相似(FSIMC)指数,提出并构建了基于一种新权重函数的去噪算法.该算法利用新的权重函数计算图像块与图像块之间的相似性,分别对RGB三通道内图像中所有的像素点进行逐块滤波得到估计值,只要平均这些估计值就得到去噪后的整幅图像.实验结果表明,相对于快速非局部均值算法,所提出的算法在PSNR和FSIMC方面均有提高,同时也保留了图像更多的细节信息.
展开▼
