【摘要】 从冗余字典中得到信号的最稀疏表示是一个NP难问题,即使是次优的匹配追踪仍然相当复杂。该文提出一种多原子快速匹配追踪算法。该算法首先将冗余字典分解成M个非相干的子字典,每次迭代分别从各子字典中至多选取一个满足条件的原子组成多原子集;最后通过求信号在多原子集上的正交投影,得到信号的多原子稀疏逼近。实验采用真实音频信号进行仿真;结果表明新的算法获得与匹配追踪相当的稀疏逼近性能,同时大大提高了信号稀疏分解的速度。更多还原

【Abstract】 Find the sparsest representation of a signal using a redundant dictionary is a NP-Hard problem.Even sub-optimal algorithm such as Matching Pursuit(MP) remains highly complex.In this paper,a multi-atoms rapid matching pursuit(MAMP) algorithm for signal sparse decomposition is proposed.Firstly,the redundant dictionary is partitioned into M incoherent sub-dictionaries.At each iteration,at most one matching atom is selected from each sub-dictionary to form a multi-atoms set.Finally,the multi-atoms signal approximation is achieved by orthogonal projection on the space of multi-atoms.Experimental results for actual speech signal show that the approximation performances of the proposed method are comparable with those of the matching pursuit.Meanwhile,the speed for signal sparse decomposition is greatly improved than those of matching pursuit.更多还原