Research On The Methods Of Computer Face Detection And Recognition

Biometrics is a kind of science and technology which using individual physiological or behavioral characteristics to verify identity. It provides a reliable and robust approach to the identity recognition. Automatic face detection and recognition is one of the most remarkable branches of biometrics as well as one of the most active and challenging tasks for image processing, pattern recognition and computer vision. It is widely applied in commercial and law area, such as mug shots retrieval, real-time video surveillance in security system and cryptography in bank and so on. Face recognition has direct, friendly characteristics and is no psychological obstacles for users. This dissertation mainly studies the approaches for frontal face detection and recognition.The goal of this paper is to explore the efficient methods in face detection and recognition, and the research work is concerned on the two aspects: face detection and face recognition. The main research works and contributions are as follows.(1) We analyzed the essential of the 2DLDA and alternative 2DLDA algorithms. In this paper we proposed a parallel feature extraction based method named 2DCCDA. In this method, horizontal and vertical features are firstly integrated, which efficiently overcome the drawback of the (2D)2LDA method where features along each directions are unbalanced extracted sequentially. Secondly the further complex feature matrix is extracted by C2DLDA. And then pick out some components from this matrix to form a feature vector according to their discriminative abilities. Compared with 2DLDA and (2D)2LDA methods, the proposed 2DCCDA can present an image with less feature components and the recognition rate can be improved. (See 6.4)(2) We analyzed various algorithms base on the Principal Component Analysis in detail, and then proposed a Binary-structure-based Feature Selection (BFS) for face recognition. The main idea of this method is to view two classes as a group, and choose the most suitable eigenfaces for discriminating these two classes. BFS method considers that different group need different eigenfaces for classification, and then chooses the most suitable eigenfaces for an arbitrary group. A certain eigenface is just employed for classification when it is really needed. Thus the recognition accuracy is greatly improved. (See Chapter 5)(3) We proposed a skin color and eye locating based method for face detection. The skin color model is firstly used to determine the face color region, on which the masks according to the face geometric characteristics are used to determine the rough region of existing face. Then the directional models based eye locating is performed. The pair of eye locations can be utilized to adjust the face location more accurately as well as verify the presence of face. The experiments demonstrate the efficiency of this method under complicated background. (See 3.3)(4) Consider the realistic situation of face recognition when only one training sample per class is available, the LDA or 2DLDA based methods will fail due to the fact that the within class scatter cannot be evaluated. In this paper, we proposed a 2DCCDA based multi oriented method for face recognition. The single training sample problem is artfully solved by sampling, and features along multi directions can be extracted. (See 6.5)...