The Research on Face Feature and Face Recognition
|School||Lanzhou University of Technology|
|Course||Detection Technology and Automation|
|Keywords||Face Recognition image processing K-L Convolution linear discriminant analysis(LDA) criterion function|
With the developing of the technology of micro-electron, computer and Internet, the traditional identification methods meet challenge, such as using passwords and IC card. The research of new method for identification is necessary. Because of the unity of human body feature, people think of identification by use of biological features. By now, the fingerprint identification and the iris identification have high recognize rate. Compared to other biological features, the face has direct, friendly and convenient character. In recently, many international researchers focus on the automatic face recognition and have done much study on itFace detection and location and face feature extraction are the two parts of the Face Recognition system, the latter is our main work in this page. And the techniques of pre-processing, feature extraction and classifier are also studied in the thesis.In the stage of image pre-processing, lighting compensation and hist equality aredone first. Principal Component Analysis (PCA) method is used to extract feature in this thesis. on the theory of Principle Component Analysis(PCA), the face image is treated as a vector and get eigenvalue vectors using K-L convolution. The linear combinations of these Eigenvectors are used to describe, represent, and approach the face image. Thus feature can be extracted. A novel linear discriminant criterion function is applied in the classifier.Fisher linear discriminant analysis often encounters a fundamental difficulty caused by singular within-class scatter matrix occurring in Small sample size problem, such as Face recognition. This paper present a new linear discriminant criterion that can find the optimal weight vector according to the criterion despite the within-class scatter matrix.In the end, we design one recognition system. We get the ideal result and our methods are proved to be perfect.