Study Based On Nuclear Technology, The Fsda Face Recognition Algorithm

At present, face recognition has become a hot topic in the research of pattern recognition and artificial intelligence. Despite we are skilled at recognizing the identities of people from their faces, but face recognition by computer automatically is a complex and difficult problem, which has referred to the field of pattern recognition, image process, psychological and physiological. Its important theory and application have always inspirited people to work on it.The aim of face recognition can be easily described as follows: Given a certain static picture or dynamic video picture of scene, try to detect and recognize one or more persons in them on the basis of prearranged face database indicating relevant ID information. In the realm of computer vision, the process of face recognition consists of three parts: face detection, feature extraction and recognition, in which feature extraction is the most significant part.This thesis studies the theories and methods of FR (face recognition) systematically, focusing on kernel technology and Foley-Sammon discriminant analysis, aiming at using them to solve the key problems in FR, such as feature extraction and classifier design.In fact, due to original sample is highly complex and non-linear, traditional methods could not achieve satisfactory result. This thesis has inspired by the kernel technology in SVM and suggested a KFSDA algorithm, which integrate kernel technology and FSDA algorithm. KFSDA algorithm uses kernel to transfer original non-linear training sample to high dimension linear feature space H, then do FSDA algorithm on it. Following is how FSDA implement: get a group optimal discriminant vectors which meet orthogonal condition when the Fisher discriminant function get max value, then projection for low dimension feature space.Using this algorithm for feature extract, we can not only make complex and non-linear sample simple and linear, but also reduce redundancy information in feature space. Finally, a practical ORL experiment using algorithm above demonstrates the efficiency.