原子簇快速匹配追踪算法  

Quickly matching pursuit algorithm based atom clusters

作　　者：崔现军[1] 赵歆波[1] 林增刚[1] 张艳宁[1] 

机构地区：[1]西北工业大学计算机学院陕西省语音与图像信息处理重点实验室,西安710129

出　　处：《计算机应用研究》2013年第2期396-398,409,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(60903126;60872145);西北工业大学基础研究基金资助项目(JC201122)

摘　　要：针对稀疏表示中匹配追踪算法计算复杂度过大的问题,提出了基于冗余字典原子相关性的匹配追踪算法。该算法利用相邻迭代过程中匹配原子的相关性对冗余字典进行簇化,得到M个多原子集合(原子簇);每次迭代过程中利用LVQ神经网络的快速学习能力从原子簇中选取目标簇;最后在目标簇中选取匹配信号结构的若干原子进行信号的稀疏逼近。实验采用一维稀疏信号进行仿真,结果表明与匹配追踪算法相比,其逼近性能相近,同时稀疏分解速度大大提高。This paper proposed a new quickly matching pursuit algorithm based on atomic correlation for sparse representa- tion. Firstly, it clustered the atoms of redundant dictionary into M multi-atoms sets after making full use of the atomic correla- tion in the iterative process. Then at each iteration, it selected one multi-atoms set as the target atom cluster. Finally,it got some atoms that met the matching conditions from the target atom cluster and the signal approximation. Experimental results for 1 D sparse signal show that the calculation speed of the algorithm increases significantly compared with MP＇ s. Meanwhile, the approximation performances of the proposed method are comparable with those of the traditional matching pursuit methods.

关 键 词：稀疏表示 匹配追踪 原子簇 LVQ神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]