Parallelizing Research And Implementation Of Face Recognition

Face recognition is the hotpot in the fields of biometrics recognition. In face recognition, the dimension of original image is pretty high. To process the original image directly will improve the computational complexity of face recognition. There are many algorithms reducing dimensions, which will loss many features. So, the dimension of face feature is restricted, which will be the bottleneck of face recognition system in speed. The parallelization research of face recognition can improve the speed without reflecting the face recognition rate.Until now, the method of face recognition is very accurate and simple. The main way to improve the speed of face recognition is to improve the algorithm. But the effect is tiny, and the speed improved is small. The multi core compute bring a chance for speeding the face recognition system. This paper speeds the recognition system through partitioning the system with threads and grouping the data.This paper implements a face recognition system based on PCA algorithm and BP neural network, which parallelized using multi threads and SSE instructions. Firstly, preprocesses the original face image. Secondly, extracts the feature by PCA algorithm. The system is completed through training and identifying the face feature. Lastly, this paper partitions the system with multi threads, and groups the data to implement the parallelization. To parallelize the BP neural network is difficult, and the speed of convergence is slow. So, this paper gets a multi BP neural network by improving the BP neural network.The recognition rate of face recognition in this paper reaches92.5%, which increases5%compared to the system using PCA only and2%compared to the system based on PCA and BP network. On the platform of Intel CoreTM2computer with two cores, the serialized training required111seconds, faster than BP neural. The speed of parallelized system using multi threads is1.77times faster than serialized system. The speed of parallelized system using SSE instructions is3.795times faster than serialized system. The speed of parallelized system using multi threads and SSE instructions is5.972times faster than serialized system.