Sparse representation classification via fast matching pursuit for face recognition

Face recognition is a widely studied pattern recognition problem. One of the most crucial components of face recognition problems is classification. Sparse representation-based classification (SRC) has been recently proposed to considerably improve the classification performance by using the compressed sensing theory. However, SRC utilizes ℓ1 minimization for recovery. Despite being optimal, ℓ1 minimization is computationally expensive, and hence, not applicable in real-time applications. In this paper, we present the Fast Matching Pursuit (FMP) which is a compressed sensing recovery algorithm that results in a recognition time that is only 4% to 10% of that of ℓ1 minimization and approximately half the time of existing related matching pursuit approaches. This significant speedup does not come at the expense of any degradation in the recognition rate.