A New Linear Appearance-based Method in Face Recognition

Human identification recognition has attracted scientists for many years. During these years, and due to increases in terrorism, the need for such systems has increased much more. The most important biometric systems that have been used during these years are fingerprint recognition, speech recognition, iris, retina, and hand geometry, and face recognition. For comparing biometric systems, four features have been considered: intrusiveness, accuracy, cost, and effort. The investigation has shown that among the other biometric systems, face recognition is the best one [1].

A face recognition system has three parts—face localization, feature extraction, and classification. In face localization, part of the background and other parts of the image that may influence the recognition process is removed from the image. For this reason, the face is found in the image and the system just works on this part of the image. For simplicity, we ignore this part of the system. In the feature extraction part, the unique patterns of the face will be extracted from the image, and in the classification part these patterns will be placed in the class in which they belong. Each class shows a person’s identity. The process of extracting the most discriminating features is very important in every face recognition system. In this chapter, we introduce fractional multiple exemplar discriminant analysis, which is a variation of a linear discriminant analysis algorithm. The results show that the proposed method, combined with RBF neural networks, has better results in comparison to other methods.