基于结构聚类的图像去噪

为了克服传统BM3D去噪算法的不足,根据图像局部结构相似性提出了基于结构聚类的图像去噪算法。首先根据均值进行粗聚类构成块群;其次利用鲁棒数据归一化构造结构相似子群;最后对子群进行去噪,如果子群容量大于1,运用BM3D对该子群进行去噪处理,反之,运用基于阈值的DCT去噪算法对该块进行去噪。实验结果表明,该算法保护了图像的结构信息,相对于传统BM3D算法提高了图像的视觉效果。

Image denoising based on structure clustering

In order to overcome the deficiencies of traditional denoising algorithm BM3D, this paper proposed denoising algorithm based on structure clustering according to the local structural similarity. First, processing coarse clustering to get block group according to the mean, followed by the use of robust data normalization to construct structure similar subgroup. At last, denoising the subgroup, if subgroup capacity is greater than one, using BM3D to denoise the subgroup, on the contrary, using DCT denoising algorithm based on the threshold to denoise the block. The experimental results show that the algorithm protects the structure of the image information and improves the image visual effects compared with traditional BM3D algorithm.