Parallelization of elastic bunch graph matching (EBGM) algorithm for fast face recognition

This paper presents a parallel method for EBGM face recognition. Compared with other methods such as principal component analysis (PCA) and linear discriminant analysis (LDA), EBGM has the advantage of higher accuracy, however, with more computational time and memory usage, which also mean less practicability. We propose a parallel method for EBGM by balancing the unit of images. We distribute the training process and allot the probing images to all processors equally, and then the recognition process is carried out in all processors simultaneously by communicating with each other. The experimental result on Message Passing Interface (MPI) platform shows that the speedup and efficiency maintain excellent with different problem size and the number of processors. Moreover, memory usage also decreases on each processor.