Face Recognition Based On CNN And Hardware Implementation

With the rapid development of computer science and Internet technology, Face recognition technology is widely used in various fields such as public safety, public security, judicial and criminal investigation, information security and access control systems. Such as in public security areas,it needs to find out some peticular criminal information in the system face database, or in the access control system to quickly identify and match the identity of the relevant information. The human face as a stable, intuitive, high degree of identification of biological characteristics attracts more and more researchers’ attention.This paper study on the face recognation under the small munber of targets. After analyzing various kinds of face recognition algorithms and their properties, selects Convolution Neural Network(CNN) in Deep Learning to realize face recognition. Firstly, this paper introduces the principle of convolution neural network , and then analyses the characteristics of classical activation functions. Some research shows that the parameters in convolution neural network are in a large amount of redundancy. When the hyperbolic tangent (tanh) is selected as the activation function, the percent of neurons activated in the network is high. while the Rectified Linear Units(ReLU) activation function has sparse activation characteristics, it has poor fitting on data in this paper’s case of less network layers due to its linear response edge. this paper proposes a new activation function (named newfuc) by combining the properties of the two activation functions, The newfuc activation function has a certain degree of sparse activation and nonlinear response characteristics. In this paper, the training results in the ORL face database, FERET face database and Yalefaces database demonstrate that in the case of high redundancy of network parameters, the newfuc activation function has higher recognition rate than the original activation function, and the redundancy of the original network parameters and the computational complexity of the algorithm are reduced due to the sparse characteristic.Due to the parallelism of the convolution neural network algorithm, hardware implementation can accomplish mutiple arithmetic units’ calculation at the same time to accelerate CNN’s calculation process. In this paper, the hardware module of face recognition algorithm based on convolution neural network is designed. Firstly, the whole algorithm is divided into control module, conv-pooling module, full-connection module and classification output module: the conv-pooling module and the full-connection module extract the face feature, the classification output module outputs the face label after the feature’s classification,and the control module coordinates the work of the whole hardware, the RTL level of all modules are described by verilog. After verifing the logical correction and synthesizing with the Quartus Ⅱ, the design is downloaded to the FPGA development board, which verifies the correctness and feasibility of the hardware implementation.