基于前向预测策略的块正交匹配追踪算法研究  

Research of Group Orthogonal Matching Pursuit Algorithm Based on Forward Prediction Strategy

作　　者：曾辉[1] 胡凯[1] 肖芬[1] 徐真珍 黄紫橙 

机构地区：[1]湘潭大学智能计算与信息处理教育部重点实验室,湖南湘潭411105

出　　处：《计算机工程》2014年第12期132-135,共4页Computer Engineering

基　　金：国家自然科学基金青年基金资助项目(61202192)

摘　　要：在块正交匹配追踪算法中,候选集匹配原子块的选择策略对信号重建具有重要作用,但是该算法在迭代选择原子块的过程中,每次选择当次最优迭代的原子块,并不能保证最终迭代性能是最优的。为此,提出一个基于前向预测的最优匹配原子块选择策略的块正交匹配追踪算法,通过预测原子块在未来迭代过程中的性能选择最优匹配的原子块。实验结果表明,与块正交匹配追踪算法相比,该算法的重建误差随前向预测参数的增大而减小。A set of candidate matching selection strategy with atomic blocks is crucial to the performance of the signal reconstruction for Group Orthogonal Matching Pursuit（ GOMP） algorithm. However,in the process of iteration selection for atomic blocks,when select the local optimum atomic block in each iteration,it cannot guarantee that the final iteration is optimal performance. So this paper proposes the GOMP algorithm based on Look Ahead（ LAGOMP） . The selection of atomic block in the current iteration depends on its effect on the future iterations. Experimental results show that, the proposed algorithm performs much better than GOMP algorithm for the reconstruction error decreases with the increases to the forward prediction parameters.

关 键 词：压缩感知 块稀疏信号 前向预测 块正交匹配追踪 局部最优原子 信号重建 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]