Research About Access Control System Based On Face Recognition

Over the last decades, face detection and face recognition has made great improvement, but, none of them solve the core problem. The reason lies in (1)face detection algorithm is time cost that many face recognition system in real world can not meet the requirement of real time. (2)high dimension of training set and small scale training set for current face recognition algorithm. (3)feature variance for face recognition of in-class is bigger than between-class some time. (4)effective of face feature description and highly precise of core recognition algorithm. (5)how to improve the robustness of face recognition to inevitable misalignment of the facial feature. In addition, the quality of camera, efficiency of algorithm, acutely change of illumination, etc. should all considered in practical system.In this thesis, all the problem presented above are studied, solutions to them are presented also. We introduce the brief history of face detection, face recognition and recent research outline firstly. Then, we explain the face detection algorithm based on Haar feature and AdaBoost classifier, meanwhile, the complexity and efficiency of algorithm are analysed. After that,, we improve the current face detection from the aspect of motion detection and edge orientation matching. Then, we give a brief introduction of Gabor filter methods and its application in face feature extraction, and present a method of Gabor-Sobel combined with LTP for the improvement. By comparison, we find that the improved methods is much smaller for in-class distance and much bigger for between-class distance, which is good for the final classification. Follow that, we introduced the theory of SVM and boosting. And we analysed deeply for how the boosting algorithm enhanced classification performance. Based on that, we try to combine the SVM with boosting which result a higher performance of classification. Besides, we presented a methods of increase virtual training samples by way of disturbing face location, which shows more robust performance.Finally, we designed an access control system based the face detection and face recognition algorithm presented above, and introduced the features of every sub system in detail.