Face Recognition Based On Gradient Feature With Dual-channel Convolution Neural Network

Face recognition is widely used in various fields,such as medicine,security verification system,public security system,human-computer interaction and so on.Face recognition under visible light illumination condition is still easily affected by variation of ambient light.However,face recognition under active near-infrared light illumination condition is robust to the variation of ambient light.Meanwhile,most recognition methods extract feature in the pixel domain,but ignore the features in gradient domain.Gradient domain explicitly considers the relationships between neighboring pixel points.Besides,in some publicly available face databases,there is a problem of insufficient sample size.Therefore,it is urgent to train an effective face recognition algorithm for small size database.In order to solve aforementioned problems,this thesis presents a face recognition method using Dual-Channel Convolutional Neural Network(DCCNN)combined with gradient domain information to recognize near-infrared 2D and 3D face.The whole network is divided into two channels: one channel uses the whole face image as the input,the other use the gradient image as the input.The convolution neural network is used to extract the features from both pixel domain and gradient domain.Finally,the geometric features of the face are fused by concatenation fusion in the full connection layer.The PolyU-NIRFD and CASIA-HFB-3D databases are used to test our method.The experimental results show that the recognition rate achieved by the DCCNN is 3%-10% higher than what other algorithms can achieve.