Fast Matching Pursuit with Multi-Gabor Dictionaries

Finding the best K-sparse approximation of a signal in redundant dictionary is an NP-hard problem. Suboptimal greedy matching pursuit (MP) algorithms are generally used for this task. In work, we present acceleration technique and implementation algorithm acting on multi-Gabor dictionary, i.e., concatenation several Gabor-type time-frequency dictionaries, each which consisting translations modulations possibly different window time frequency shift parameters. The based pre-computing thresholding inner products between atoms updating residual directly coefficient domain, without round-trip to domain. Since proposed involves approximate update step, provide theoretical experimental results illustrating convergence resulting algorithm. written C (compatible with C99 C++11) also Matlab GNU Octave interfaces. For some settings, up 70 times faster than standard Matching Pursuit Toolkit (MPTK).