Sparse Matrix Recovery from Random Samples via 2D Orthogonal Matching Pursuit

Since its emergence, compressive sensing (CS) has attracted many researchers’ attention. In the CS, recovery algorithms play an important role. Basis pursuit (BP) and matching pursuit (MP) are two major classes of CS recovery algorithms. However, both BP and MP are originally designed for one-dimensional (1D) sparse signal recovery, while many practical signals are two-dimensional (2D), e.g. image, video, etc. To recover 2D sparse signals effectively, this paper develops the 2D orthogonal MP (2D-OMP) algorithm, which shares the advantages of low complexity and good performance. The 2D-OMP algorithm can be widely used in those scenarios involving 2D sparse signal processing, e.g. image/video compression, compressive imaging, etc. Index Terms Compressive Sensing, 2D Sparse Signal, Recovery Algorithm, Orthogonal Matching Pursuit. This research was supported by National Science Foundation of China (NSFC) (grant no. 61001100) and Provincial Science Foundation of Shaanxi, China (grant no. 2010K06-15). The author is with the College of Information Engineering, Northwest A&F University, Shaanxi Yangling 712100, China (email: [email protected]). February 1, 2011 DRAFT IEEE TRANSACTIONS ON SIGNAL PROCESSING (SUBMISSION) 2