Shape Index Feature Based Face Detection And Recognition

Face Recognition is a kind of biometric identification technology which is based on facial information. This technology has been widely used in today’s society, such as:person identification, video surveillance, access control system, face AF system, photo gallery. The complete face recognition technology including face detection, face alignment, feature extraction and feature classification. Currently, the face recognition technology has been studied for more than30years. However, until today, uncon-strained face recognition is still a very difficult problem, since the same person may look very different under different lighting, expression and poses. Therefore, the study of accurate and rapid face recognition under natural conditions is a both practical and challenging work.This paper proposed the use of shape index feature to solve the face pose problem in face detection and recognition. The algorithm first determines the position of several landmarks, such as eyes corner, mouth corner and nose tip, with face alignment tech-nology. Then we establish a local coordinate system centre at each landmarks. Finally, we extract feature in these local coordinate systems. The shape index feature takes full advantage of the internal face structure information, and ensure to extract feature in the correct position when the face pose and expression change. Thereby, it improves the discriminate power of features.In addition, in order to use the shape index feature in face detection task. This paper proposed a joint cascade model to handle classification and regression problem simul-taneously. The algorithm distinguish faces and non-face area, and predict the position of each landmarks at the same time. It combines face detection and face alignment in a single task.This paper also proposed shape index high dimensional feature and joint bayesian model, which achieves high accuracy in face recognition task. In order to allow us to better apply our face recognition algorithm into practical, we also propose a rotated sparse regression algorithm. It significantly reduces the computational and storage cost when mapping high dimensional face feature into low dimension with little accuracy loss.We evaluate our algorithm in various databases. All the experimental results show that our algorithm can effectively improve the accuracy of face detection and recogni-tion. Among them, our face detection algorithm also achieves the first place on FDDB and AFW dataset.