Optimization Method for Parallel Algorithm for Face Recognition in Graphic Images

The aim of the research is to study the effectiveness of high-performance Symmetric Multiprocessing (SMP) systems. The criterion for effectiveness is the running time of the face recognition algorithm. The principal component analysis (PCA) as well as methods for parallelization and optimization of the PCA algorithm are described. The main part of the algorithm that takes the most time is parallelized and optimized for CUDA. A new mathematical model is used, which increases the speed of the algorithm. Also, work was done to find the optimal strategy for using CUDA kernels. Result: The face recognition algorithm based on the PCA is built and tested. Efficiency of the approach is illustrated with examples.