基于K奇异值分解的多级图像去噪算法

MULTI-STAGE IMAGE DENOISING ALGORITHM BASED ON K-SVD

针对一般去噪算法对信号的利用度不够的问题,提出一种基于KSVD的多级去噪算法。该算法通过加强观测图像及利用残差信号相结合的方法对图像进行联合去噪,在一定程度上增加了可恢复的细节信息。同时考虑到在噪声稍高的环境中,噪声原子极易污染字典,所以算法利用Bartlett校验来达到对噪声原子监测的目的,并通过替换等手段合理地优化字典。实验结果表明,该算法能够显著地提升去噪表现,尤其是对于富含纹理的图像有着更好的表现,不仅峰值信噪比更高,细节部分也表现得更加优秀。

Aiming at the problem that the general denoising algorithm fails to make full use of all signals, we propose a multi-stage denoising algorithm based on K-SVD. The algorithm enhanced the observed image and used the residual signal to jointly denoise the image, so as to increase the recoverable details to a certain extent. Considering that noise atoms were easy to pollute the dictionary in a slightly noisy environment, it used Bartlett verification to achieve the purpose of monitoring noise atoms, and reasonably optimized the dictionary by means of replacement and other means. The experimental results show that the proposed algorithm can significantly improve the performance of denoising. For images rich in texture, it has better performance, such as higher peak signal-to-noise ratio, better performance in details.