Research And Implementation Of Face Detection And Recognition Based On Faster R-CNN

With the rapid development of information technology such as computer,Internet,mass storage and so on,the status of identity authentication technology is more and more prominent in the present society.The role of information security is becoming more and more important,especially carry out identity authentication quickly and effectively.Face detection and recognition is one of the important ways of identification technology.The method of deep learning advanced by professor Hinton has been paid great attention by the academic community,and more and more scholars have used the depth learning to solve the problem of face detection and recognition.The convolution neural network model in depth learning is the most commonly used model in human face detection and recognition.In this thesis,a CNN model is constructed by analyzing the CNN model's image characteristics.The model is improved on the basis of traditional VGG model.The three aspects of innovation of this thesis is as follows.Through the intensive study of the Fast R-CNN algorithm,it is found that in the RPN network model,exhaustion has been done on the last layer convolution feature by sliding the window,but it produces large redundancy because the size of the sliding window is fixed.Therefore,in this thesis,a modified RPN network(IRPN)is proposed by using the fixed size segmentation strategy instead of the sliding window,and the detection performance of the face has been improved obviously.VGG network requires fixed size of the input image size.If the size of the face image does not meet the requirements of convolution neural network,you need to scale the face image,but scaling processing will cause the loss of image space information in a certain degree.After studying this problem,the reason has been found: it is because the dimension of the fully connection layer must be fixed.So in this thesis,the space pyramid pooling technology is introduced in the VGG network to improve the accuracy of the VGG model for face recognition.In this thesis,a set of face detection and recognition software based on Faster R-CNN been designed and implemented.The testing of the software shows,the rate of face detection and recognition can be more higher,at the same time,the software has strong robustness for faces with different gestures and expressions.Experimental results show that the accuracy rate of face recognition based on improved VGG model is 99%.