Public Safety Oriented Silent Liveness Detection And Mask Face Recognition System

With the development of computer vision and image recognition technology,face recognition technology is also becoming more mature,which is applied to various scenes such as daily attendance,payment,access control,and punch card.However,these scenes are vulnerable to the intrusion of forged faces and will pose a threat to people’s property safety.With the outbreak of COVID-19,people are required to wear masks in public places.Face recognition is also an important research content in today’s society,which is related to people’s health and safety.At present,there are some disadvantages in living detection,such as low accuracy,easy to be affected by the environment,and requiring users to cooperate with tedious actions.At present,mask face recognition technology requires users to upload the sample images of the face with masks,which is cumbersome and has the problem of lack of samples.Aiming at the difficulty of living detection,this paper proposes SVM algorithm based on improved LBP texture features and improved vgg-16 neural network algorithm;aiming at the problem of mask face recognition,this paper proposes an automatic generation algorithm of batch mask face samples.Then the models are compressed and integrated into the embedded system.(1)Research on mask face recognition algorithm.In order to solve the problem of insufficient samples in mask face recognition,this paper studies the face key points location algorithm based on multi-level cascade regression tree to locate the key points,and automatically generates mask face samples in batches according to the location points.In this paper,by comparing various face similarity calculation methods to select the appropriate algorithm to compare the mask face,so as to realize the face recognition of wearing masks.(2)Research on SVM live detection algorithm based on face texture features.In this paper,LBP histogram is used to extract outliers from living images,and then LBP histogram is used to extract outliers Finally,SVM(support vector machine)is used to train these feature vectors and save the SVM model based on texture features.(3)Research on face detection algorithm based on convolution neural network.Traditional live detection requires the user to cooperate with the camera to make action to identify.In order to achieve high-precision silent live detection,this paper studies the improved vgg-16 neural network,adds the squeeze exception module in the appropriate position of the network,and integrates batch normalization and random deactivation layer to simulate whitening operation,so as to improve the generalization performance of the model.Finally,a series of simulation experiments are carried out The model is easy to be transplanted to the mobile terminal to complete the deployment of the mobile terminal.(4)The realization of face detection and mask face recognition system.In order to integrate the advantages of the two algorithms proposed in this paper,the system captures multiple images and calculates the average of the recognition results of the two algorithms on multiple images to get the final result.Finally,the system is deployed on Android development board,which is a face detection and mask face recognition system based on mobile devices.Finally,the system is tested.The experimental results show that the system can quickly and accurately detect whether the face is living or not,and realize face recognition.