A Fast and Simple Face Detection Algorithm Using Neural Network and Its Implementation on FPGA

Abstract

Face recognition is one of the interesting types of biometric which determines the presence or absence of human faces in the picture. In this paper, a face recognition system is presented that benefits from an optimized architecture based on the MLP neural network. The proposed method considerably improves the speed and the accuracy of detection compared to traditional architectures of neural network. To reduce the overall computation, neural network is organized so that to be able to rule out the majority of the non-image areas located in the image’s background before applying the main algorithm. An important advantage of this new architecture is its homogeneous structure that makes it suitable for optimized implementation on a hardware platform. In this work, FPGA is used as the platform for implementation of the proposed algorithms. The implementation was done considering Taylor expansion of the activation functions. The performance of the proposed method and the implemented system was evaluated on the BioID dataset. Accomplishment of the proposed method is high precision while reducing training time and total calculations, together with appropriate robustness. Finally, a comparison with other face recognition methods has been done to show the performance of the presented system. The comparison result shows that the proposed system outperforms the other mentioned methods.