The Research&Development Of Face Recognition In The Wild

With the development of the Internet and mobile Internet, the rapidly growing popularity of smart mobile devices has brought new opportunities for face recognition technology. Besides the traditional security services such as access control, video surveillance and telecommunication, the face recognition technology will be uesed in many new scenes due to the development of camera in mobile phones and computers. In these new scenes, the facial images which need to be identified are from totally uncontrolled environment for instance, images from surveillance cameras, or from the Web. The numerous variations of a face image, due to changes in lighting, pose, and facial expressions have brought great challenges in face recognition technology.This paper introduces the main algorithms and techniques of face recognition in the wild. We focus on the two main kinds of face recognition tasks:face identification and face verification.The steps in our face recognition systems include face detection, key points detection, geometric correction to solve the pose and facial expressions, optical correction to sovle the impact of lighting, extract SIFT features in key points, get advanced features through the Simile classifications and finally get the verification or identification result through the high-level SVM classifier. In this paper the author has constructed many useful face image datasets, which are used in experimental testing, training Simile classification, geometric correction, and data mining. Besides the author focuses on the face image preprocessing, advanced feature extraction and face application. The face recognition system is designed and implemented through theoretical analysis and experimental results.For face verification, our method is tested on the state-of-the-art dataset, the Labeled Faces in the Wild(LFW) and the accuracy we can get is91.78%±1.35%. The true positive rate/false positive rate is0.93/0.1, which ranks top on the LFW official web site, proving the feasibility and robustness of system. For face identification, we test on two sub-datasets of the Orange Face Dataset (OFD). The accuracies are all over95%on the two datasets, proving the expandability of system.