Sparse approximation using fast matching pursuit

Matching pursuit based on geometric dictionary has shown to be a powerful tool for sparse image representation. The main obstacle to its application in real world is the computational complexity. In this paper, a modified algorithm is presented to address this issue. The dictionary with anisotropic refinement atoms is used to provide the approximation ability. Meanwhile the pursuit implementation is significantly speeded up by employing both sequential and parallel techniques. Experimental results show that compared to the latest matching pursuit approach, the proposed algorithm offers a speedup of 27.7-36.7 while maintaining the approximation quality. It is very promising for flexible image coding at low bit rate.