中文摘要：

在基于压缩感知的正交匹配追踪算法中，候选集原子的选取对最终的重建性能至关重要．文中结合前向预测和回溯两种策略更新候选原子集，提出了一种基于预测与回溯的正交匹配追踪（LABOMP）算法．该算法通过设定阈值将所有迭代划分为前后期，在迭代前期，通过预测原子在未来迭代中的性能选择最佳原子；在迭代后期，加入回溯策略，每两次迭代淘汰一个前面错误选择的原子．实验结果表明：LABOMP算法是实用有效的，由于加入回溯策略修正了预测算法IAOMP的不足，使迭代后期高斯稀疏信号与二值稀疏信号的精确重建概率较LAOMP算法分别平均提高了12．5％、18．2％．

英文摘要：

In the orthogonal matching pursuit （OMP） algorithm based on compressive sensing, the selection of can- didate atoms is very important to the final reconstruction performance. In this paper, a look ahead and backtracking- based orthogonal matching pursuit （LABOMP） algorithm is proposed by combining look ahead procedures with the backtracking strategy to update a candidate atom set. In this algorithm, all the iterations are divided into early and upper stages. At the former, the optimal atom is selected by forecasting its final performance in the future itera- tions, while at the latter, the backtracking strategy is introduced, and a previous wrongly-selected atom is then eliminated once per two iterations. Experimental results show that the proposed LABOMP algorithm is applicable and effective; and that its average exact reconstruction probability for Gaussian or binary sparse signals is respec- tively 12.5% or 18.2% higher than that of LAOMP （ Look Ahead Orthogonal Matching Pursuit） algorithm at the upper stage because the backtracking strategy overcomes the disadvantage of LAOMP algorithm.