Research On Face And Id Photo Verification Based On Deep Feature Learning And Classifier Correlation Ensemble

With the emergence of large-scale tagging data and the rapid development of deep learning technologies,automatic face recognition technology has achieved great breakthroughs and even achieved performance beyond human.At present,this technology has been widely used in security,finance,monitoring,surveillance and other real life scene.This thesis studies the face recognition problem by considering a specific problem known as face verification between ID photos and daily photos.In order to solve the problems in face verification,especially for facial deep feature extraction and classification,our research was conducted from easy to difficult.A system for face and ID photo verification is built up,and based on facial feature extraction method,combined with ensemble learning method,a CCA-based classifier association ensemble algorithm is proposed to improve the performance of face recognition and its generalization ability.The main work is shown as follows:(1)Mainstream algorithm for face detection is studied.Cascaded Boosting face detection framework is improved because of its better detection performance and realtime performance.This model is trained by selecting five face modes and sixteen Harr features in order to improve the sensitivity to pose.This method not only improves the speed of face detection,but also takes accuracy and robustness of Adaboost algorithm into account.Based on ESR model,this thesis proposes an improved ESR method by voting,which realizes the function of face alignment.This voting method improves the face alignment accuracy of the model.(2)In this thesis,CCA-based classifier association ensemble algorithm is proposed.This ensemble learning method is proposed by combining classifiers pruning and weighting together to achieve better classification performance and stability.In the first stage,a sparse regression method is proposed to prune base classifiers so that each test data point will dynamically select a subset of classifiers to form a unique classifier ensemble,to decrease effects of noisy input data and incorrect classifiers in such a global view.In the second stage,the pruned classifiers are weighted locally by a fusion method,which utilizes the generalization ability of pruned classifiers among nearest neighbors of testing data points.By this way,each test data point can build a unique locally weighted classifier ensemble.Analysis of experimental results shows that the classification results of our method are better than other ensemble methods such as Random Forests,Majority Voting,AdaBoost and so on.(3)Based on above algorithm,a system of face and ID photo verification is designed and implemented,which can match the photo with the real face.By operating system,result indicates that it can be used in face and ID photo verification.Moreover,this system functions well with a friendly interface,maintainability and good recognition speed,which verifies the effectiveness of the proposed method.