Research And Improvement Of Method In Low Dimension For Face Recognition

With the development of information processing technology, Compared with fingerprint and iris biological features face recognition has direct, friendly and convenient character. So that face recognition technology will become the most potential method of identity recognition. It is attractive in pattern recognition, image processing and computer vision.In this paper, collecting and learning many essays, researching papers related human face detection and recognition of the domestic and international in recent years, discussed some methods about face detection and face recognition. In the foundation of summarizing the existing methods of face examination, feature extraction and recognition, we propose an improved method of face recognition.Face recognition includes three parts: face detection and localization, feature extraction and classification. This thesis presents the studies of face detection and feature extraction in face recognition.Firstly we introduce some typical color model firstly in this paper. The distribution and characteristic of skin color is firstly analyzed in color model. We explore the clustering characteristics on xanthous face, and point out that HSV color model can be well used to express the clustering characteristics of xanthous faces. Then, we study on the distribution character of complexions of human face in HSV color models, set up an appropriate complexions model suitable for xanthous face, and propose a good detection algorithm on the human face features. Abandoned carrying on the method to gradation image processing, we authenticated the human face through the method of the S color component space threshold value division. After carry on the experiment to many different pictures and photograph, the experiment result finally indicated that this method effectively enhance accuracy, betimes in human face detection. In this article, we introduced the principle and the realization of the K-L algorithm, the Principal Components Analysis (PCA) and the Linear Discriminant Analysis (LDA) in detail, and discussed an improvement method in the foundation the Linear Discriminant Analysis: in the lower dimension space, the inside variable of the example lager than the between variable because of the outside factors such as illumination condition and the facial expression, which causes the error rate to LDA method higher specially using Euclidean distance. In view of this problem, we advanced a method in low dimension space for face recognition based on LDA method, improved the limitation of LDA in the lower dimension space, combined with the discriminatory power, and formed a kind of improvement the LDA algorithm. The theoretical analysis and the experiment result proved that this algorithm strengthens the person face recognition, improved the recognition rate in the lower dimension space, and enhanced algorithm practicability.We emphases discussed the nearest neighbor classifier and support vector machine (SVM) based on the statistical study theory.In the paper, we use nearest neighbor classifier to carry the experiment through the ORL person face database. The result indicated: the method has a higher recognition rate than the traditional LDA and PCA in the lower dimension space.