Research And Application Of The Practical Face Recognition System

The attendance management of the vast majority of primary and secondary schools in our country is still relying on the teachers' naming way.This kind of traditional naming way takes long time and performs inefficiently.Besides,the parents are not able to get the information about their children's entering and leaving campus in real time.Which leads them to be lack of effective supervision of their children.Under this background,in order to solve the above problems,this group has researched and implemented an attendance management system focusing on the students' entering and leaving campus.This system collects photos relying on the card reader terminal when the students swipe their cards.Then some related information including photos is pushed to the parents through the WeChat which enable parents to supervise their children's entering and leaving campus.The school administrators can quickly find out who is absent by the data retrieval technology and prevent the students' cheating behaviors by the face recognition technology.With the help of the system,the efficiency can be improved greatly.This paper describes the attendance management system based on face recognition which is a part of the whole system.The main works done includes the following points:(1)The Haar-like feature and the AdaBoost algorithm were used to train the face detection classifier which can detect the face area quickly and accurately.The detection time is a few hundred milliseconds and the accuracy rate is up to 96.51%.The SIFT feature and the SDM algorithm were used to implement the facial landmark detection.Then face images were scaled,rotated and cut out according to the distance and the angle of the two eyes to complement face alignment.(2)The 400 dimensional feature vector which was obtained by reducing the dimension of the HD-LBP feature vector with PCA algorithm was selected to represent the face.Then the cosine distance between the vectors was a measure of similarity between faces.The best threshold for face matching verification was set by SVM algorithm.The accuracy rate of 84.87% was achieved by testing on LFW data set.(3)Many algorithms were implemented easily with a large number of image processing functions provided by the OpenCV.The attendance management system was implemented with the Windows API packaged in MFC and its application framework.This system solved the problem of inefficiency for school administrators and put an end to cheating behaviors.Besides,it can work with the program of WeChat public service to provide attendance information for parents.The innovation of this paper is described as follows.The HD-LBP feature which was improved based on LBP feature has been implemented.The accuracy rate has increased by 4%.Besides,both the number of scales and landmarks in the process of HD-LBP feature extraction were studied and analyzed deeply.These two parameters were optimized to balance between the computational complexity and the accuracy rate.Aimed at the problem that it is difficult to be implemented when using the PCA algorithm to reduce the dimension of HD-LBP features because of the great computational complexity,an improved PCA algorithm was proposed.It solved the problem by calculating the eigenvalues and eigenvectors of a simple matrix which has the same eigenvalues and the related eigenvectors with the covariance matrix.