Research On Face Recognition System Based On Convolutional Neural Network

Face recognition is an important branch of pattern recognition and computer vision.People pay more and more attention to the study of face recognition.It is widely used in various fields,such as human-computer interaction,security monitoring,identity recognition,skynet system,human traffic monitoring and so on.Compared with other biometrics,face recognition is a kind of biometrics with high accuracy,fast recognition speed,broad acceptance and low recognition cost.With the continuous development of deep convolutional neural network,face recognition has shown super strong performance in the field of image classification.Moreover,with the continuous improvement of structure and layer number,great progress has been made in face recognition.In this paper,a new face recognition algorithm is proposed to solve the problems existing in face recognition algorithms,such as weak robustness and complex computation,and the following research work is completed:(1)Shuffle-channel convolution structure is used to replace the original convolution structure of MTCNN,which increases the detection speed and improves the detection performance on the premise of ensuring that the detection recall rate does not decline.In addition,before the detection of Pnet in the network model,a median filter is used for rapid denoising to reduce the error detection of Pnet.At last,according to the output of the model,the appropriate Min Size value is matched with the image,and the minimum face size is dynamically modified.(2)Triplet loss structure was introduced into the Res Net network,and a new face recognition algorithm was proposed.By using Resnet to ease the advantages of the difficulty of convergence and tuning in deep networks,the Softmax loss function is removed and the Triplet loss function is used to solve the problem of the slow convergence speed of the Triplet loss function.(3)This paper solves the problem of small samples by combining data enhancement and online hard case mining techniques.If there are not enough samples of a single category in the training set,the problem of sample imbalance will occur because of thesmall number of the same category when training with the Triplet loss function.Through the use of data enhancement methods to increase the number of samples in the database,and the use of online hard case mining technology during training,the positive and negative sample pair asymmetry is eliminated,thereby achieving the effect of increasing the data set and completing the solution to the problem of small samples.(4)Based on the face recognition algorithm proposed in this paper,the design and implementation of a face intelligent small program is completed.The main purpose of this algorithm is to package it into an interface through Flask and apply it to a small program of face recognition We Chat designed by ourselves.