Two Dimensional Face Recognition Based On CNN

In recent years,face recognition has become one of the hot research subjects in several areas,such as pattern recognition,computer vision and so on.According to the types of test data,face recognition can be divided into two types: the face recognition in controlled scenes and the face recognition in uncontrolled scenes.In the controlled scenes,face recognition technology has achieved a satisfying effect.In the process of practical application,the performance of face recognition sharply declined and cannot meet the needs of practical applications when it is affected by a series of uncontrollable factors,such as illumination changes,attitude changes,occlusion and facial expression changes,etc.Since these factors that can cause the intra-class variation of the human face image in the uncontrolled scenes is much larger than the inter-class variation.The purpose of this study is to explore the face recognition in uncontrolled scenes.The solutions of key issues and problems are presented.In this article,we take into account the various kinds of interference factors and the phenomenon of "missing data" in the uncontrolled face recognition,and we also study and design the face recognition algorithms.As for the various kinds of interference factors in the uncontrolled face recognition,a novel deep convolutional neural network(CNN)for face recognition is designed in this paper.This CNN model has a number of hidden layers,which can eliminate the influence of different kinds of complex interference on the face recognition problem in uncontrolled state by the way of layer by layer extraction.As for the "missing data" phenomenon caused by the posture or the shades and other factors,the whole connection layer of the CNN model is designed in this article to connect with both of the convolution layer and the pool layer.In view of the characteristics that the more layers of convolutional neural network there are,the wider the field vision is,this kind of connection mode can both deal with local features and global features.The integration of global and local features can better cope with some of the data problems.In this paper,we simulate the algorithms referring to face database of Youtube Faces,and a face database of 157 subjects,then compared it with the artificial design feature recognition methods and other deep learning methods.The experimental results show that the proposed algorithm is superior to other artificial design feature recognition methods.The verification rate of 95.7% is obtained on the 157 face database.It also shows that the CNN model proposed in this paper has a good robustness to the face recognition in uncontrolled condition.