摘要
提出了一种结合噪声分布先验知识的稀疏表示混合去噪算法。该算法通过自适应中值滤波器进行初始化来分析噪声分布先验,对稀疏编码中的原子进行自适应加权。然后以当前原子集的极值为基准调整选取阈值,对稀疏编码中的原子进行选择淘汰。本算法避免了传统混合去噪算法的两相检测策略,时间复杂度显著降低。实验表明本算法在峰值信噪比PSNR和去噪效率上都有明显优势。
We propose a mixed denoising algorithm based on sparse representation and prior knowledge of noise distribution. The proposed algorithm utilizes the Adaptive Median Filter (AMF) to initialize and analyze the prior knowledge of noise distribution, and adaptively weight the sparse representation atom vector at the stage of sparse coding. Then, the selection threshold is adaptively adjusted by the extreme value of the current set of atoms so as to do selective elimination on atoms. Because of a avoidance of the traditional two-phase mixed denoising strategy, the proposed algorithm gains much better PSNR and faster speed.
出处
《计算机工程与科学》
CSCD
北大核心
2015年第10期1917-1923,共7页
Computer Engineering & Science
基金
国家自然科学基金资助项目(61402053)
湖南省教育厅优秀青年基金资助项目(12B003)
湖南省交通厅科技资助项目(201334)
2015年湖南省研究生科研创新资助项目(CX2015B369)
关键词
混合去噪
稀疏表示
自适应中值滤波
原子加权
mixed denoise
sparse representation
adaptive median filter
weighted atoms