Research Of Multi-pose Face Recognition Based On Convolutional Neural Network

With the continuous improvement of scientific level,face recognition technology has been widely used in security,attendance,entrance guard,ATM and so on.At present,the research on face recognition algorithms has also been extended to face recognition in complex environments.Posture change,face occlusion and age change are the main factors affecting the recognition performance of the face recognition system.Compared with traditional face recognition,convolutional neural networks directly act on the original image and have the characteristics of local receptive fields,weight sharing,and subsampling.In addition,the convolutional neural network is invariant to the panning,scale zooming,and tilting of the image.This paper has carried out research on applying the characteristics of convolutional neural networks to multiangle facial recognition systems.First,the composition and characteristics of the traditional neural network are studied,and then the structural characteristics of the convolutional neural network are studied.The size and number of the convolution kernels of the classic convolutional neural network model Le Net-5 are improved to improve the ability to extract features.Improve the pooling method of the pooling layer and reduce parameters and calculations.The regularization method is added to the convolutional neural network to reduce overfitting and enhance the generalization and sparseness of the network.Secondly,the feature dimensions extracted from the improved convolutional neural network are higher,the PCA algorithm is used to reduce the dimensions of the extracted feature vectors,and then generate feature vector libraries of different dimensions.Furthermore,the cosine similarity measurement algorithm is used to identify the target face.Finally,simulation experiments are carried out on the multi-angle face recognition method in this paper.The experimental results in the CAS-PEAL face database show that the improved convolutional neural network face recognition algorithm effectively improves the recognition accuracy rate and recognition time under multi-pose conditions.