Dictionary Learning with Rapid Orthogonal Matching Pursuit

Orthogonal Matching Pursuit (OMP) has proven itself to be a significant algorithm in image and signal processing domain in the last decade to estimate sparse representations in dictionary learning. Over the years, efforts to speed up the OMP algorithm for the same accuracy has been through variants like generalized OMP (g-OMP) and fast OMP (f-OMP). All of these algorithms solve OMP recursively for each signal sample among 'S' number of samples. This algorithm, rapid OMP (r-OMP), runs the loop for 'N' atoms, simultaneously estimating for all samples, and, in a real scene since N<<S, the proposed approach speeds up OMP by several orders of magnitude.