Hybrid MPI/OpenMP implementation of PCA

Abstract

Most surveillance systems depend on fully automated face recognition applications. The main concern is achieving high accuracy in real time. Principle Component Analysis algorithm is used for reducing the number of variables and getting the maximum variance between low dimensional data. The proposed approaches focus on data partitioning to minimize the execution time of the algorithm by distributing data over a cluster with parallel computing architecture. The first approach achieves 2975X and 102X relatively faster than the sequential implementation in the training and recognition phases, respectively. However, the second approach achieves 74X relatively faster than the sequential implementation in the recognition phase.