Research On Face Recognition Based On Nonparametric Discriminant Analysis

Biometric identification technology has become an important research direction in the area of authentication and certification. As one of the technologies of Biometric identification, face recognition has been paid much more attention because it has several advantages over other technologies, such as convenient in use, easy in acceptance etc. In this paper, a deep study on the facial features extraction methods of PCA and LDA is made. What we mainly studied in this thesis is given as follows:1. The basic principle of the methods of PCA (Principle Component Analysis) and LDA (Linear Discriminant Analysis) are introduced, and the disadvantages of the method of PCA for face recognition are analysed. Based on it, the LDA face recognition methods is realized. Through a large quantity of experiments, it shows that the method of LDA does not apply to non-linear data, and be unable to extract the non-linear structure embed in image space.2. Due to the disadvantages of failure to express the non-linear structure in the image space effectively, the method of face recognition based on NDA (Nonparametric Discriminant Analysis) is proposed. With using the idea that the non-linear structure can be regarded as a linear one in local area, the between-class scatter matrix and within-class scatter matrix are redefined, and the optimal projection direction is get by using the rule of maximizing the between-class scatter and minimizing the within-class scatter matrix, thus the features abstracted by projection not only maintain the characteristics of local structure inside class, but also gain a better compactness in class and better classification performance between classes.In addition, the number of features that NDA extracted will not be confined by the number of classes.The experiments on the ORL and AR face databases confirmed the effectiveness of the algorithm.3. A real-time face recognition system based on NDA algorithm is achieved. A detailed description of the design principle, criterion, structure and flow chart are given and analyzed.