Adaptive Sparsity Matching Pursuit Algorithm for Sparse Reconstruction英文清晰版.pdfVIP

IEEE SIGNAL PROCESSING LETTERS, VOL. 19, NO. 8, AUGUST 2012 471 Adaptive Sparsity Matching Pursuit Algorithm for Sparse Reconstruction Honglin Wu and Shu Wang Abstract— This letter presents a new greedy method, called where denotes the vector norm and is Adaptive Sparsity Matching Pursuit (ASMP), for sparse solutions -order restricted isometry constant. When , the of underdetermined systems with a typical/random projection RIP implies that the projection operator nearly maintains the matrix. Unlike anterior greedy algorithms, ASMP can extract information on sparsity of the target signal adaptively with a distance between each pair of -sparse signals [3]. well-designed stagewise approach. Moreover, it takes advantage Various efforts have been made on the fast numerical of backtracking to refine the chosen supports and the current methods for sparse reconstruction with reliable accuracy and approximation in the process. With these improvements, ASMP robustness to noise. Among the existing recovery algorithms, provides even more attractive results than the state-of-the-art two major approaches are -minimization and greedy algo- greedy algorithm CoSaMP without prior knowledge of the spar- sity level. Experiments validate the proposed algorithm works rithms. -minimization methods, such as Basis Pursuit (BP) well for both noiseless signals and noisy signals, with the recovery [4] and its optimization algorithm GPSR [5], solve a convex quality often outperforming that