An Automatic Face Recognition System Based on Support Vector Machines

• Main Authors: Wang, Chun-Kai, 王君楷
• Other Authors: Tsai, Chau-Ren
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/24652885626161143290

碩士 === 國立臺灣科技大學 === 電機工程系 === 91 === The study of face recognition has been proceeded for two decades. Though many face recognition techniques have been proposed and have demonstrated signification promise, the task of robust face recognition is still difficult. In this thesis, we used the kernel-based support vector machines (SVMs) for face recognition and the subspace density estimation method for face detection to build an automatic face recognition system. In past research on face recognition, it can be find that the distribution of face images, under a perceivable variation in pose, facial expression, illumination, is highly complex and non-separable. It may cause serious performance degradation for most existing face recognition system, such as Eigenface or Fisherface. In order to overcome this problem, we first use the direct linear discriminant analysis (D-LDA) to extract the features from face images; then the kernel-based SVMs are used to classify the features with complex distribution. Furthermore, for the sake of making the system detect human faces robustly, we propose a heuristic strategy based on maximum likelihood estimation (MLE) principle which can estimate the parameter of the distribution density function of face images in non-principal subspace. By the use of the proposed heuristic strategy, our system has the capability of detecting that be robust for various facial expression, pose, obscure image, and partial occlusion. Moreover, it achieved over 90% recognition rate when tested on a database contained 50 distinct persons with variation in facial expression, pose, the size of face, illumination.