Independent component analysis is a kind of efficient technology of blind signal processing recently. Satisfactory independence is obtained from statistically separated signal. There are widely applications in the fields of signal recognition to audio, image and medical signal etc. Now it become hotspot of research in the fields of signal procession and artifical neural network.With the improved of our safety consciousness and development of computer version and intelligent pattern recognition, face detection and face recognition become research topics in computer science and more attention is paid for this research. Automation face recognition consists of three parts: face detection, feature extraction and classification. Basic theory and improved algorithms of face detection and recognition are emphasized in this paper, in which modified kernel ICA method is focused on to realize face recognition.Firstly, we introduce the relative research history, research progress and the research trend.Secondly, technology of combination of preliminary detection and precise location is used to extract section of face from image, in which modified AdaBoost method is employed to find the approximate interesting position of face, and then method of pattern matching is chosen to locate the precise face. Result of experiments show that the method can result in quick, exact and efficacious recognition.In the end, general algorithms based on statistical subspace method are studied, in which we emphasize the ICA algorithm which has the good property of local feature behave. To improve the adaption to nonlinearity, a efficient modified kernel ICA algorithm is present to realize feature extraction and recognition. It has encouraging capability of self-adaption and satisfactory performance of recognition in the experiments on ORL and Yale databases. |