Study Of The Technology Of Human Face Recognition With Computer

The technology of human face recognition is an active subject in the field of pattern recognition. There are broad application in the fields of laws, business and safety systems ect. For the particularity of human face images, face recognition with a computer is a very difficult problem and there are still many works to do before such technology can be used wildly. With the devolpment of the society, the application of face recognition systems will be wilder and brings much challenge to the reseachers.The human face recognition system is a kind of pattern recognition system based on information processing. It can be divided into two parts: feature extraction and pattern classification. The first part is to find out a set of features that can represent the images from different persons; the second part classifies the features got from the first part. The performance of the system depends on both of the parts.For the first part of the system, the methods used in the thesis are principle component analysis (PCA), Fisher linear discriminant (FLD) and independent component analysis which are statistical methods and are widly used in signal processing and pattern recognition.For the second part of the system, SVM multi-classifers are applied to classify the features from the first part. Because an SVM can only slove a two classes problem, for a K classes problem, a set of SVM must be used. How to combine a set of SVM into an efficient multi-classifier is a problem worthy of doing much research work. Althought there are kinds of traditional ways to combine a set of SVM, accoding to the analysis in the thesis none of them is efficient for a K classes problem. The method of error-correcting output codes (ECOC) is introduced to combine a set SVM into a multi-classifier in this thesis, with this method, we can use medium number of SVM to construct an efficient multi-classifier. Because of the predefined ability of error correcting, the multi-calssifer can give proper result even though some SVM give wrong decision. The complexity of the system is reduced and the recognition rate of the system is improved.