The Research Of Face Recognition Based On Two Dimensionality Principal Component Analysis

Face recognition is a technology which analyzes the person picture and extracts the effective identification informations for identifying the person status by computers. Because of the particularity of face images, face recognition problem is a hard problem in pattern recognition, it is also a challengeable problem and covers knowledge of many subjects such as signal processing, intelligence control, pattern recognition, machine vision. Face recognition consists of three parts: preprocessing, feature extraction and classification. In this thesis, we start from that three parts:(1) In this paper, we preprocess the pictures using the Gabor wavelet. First of all, study the basic knowledge of Gabor wavelet; Secondly, select appropriate Gabor filter and design Gabor filter; Finally, filter pictures using the above designed Gabor filter to reduce the impact of light, expression and so on.(2) Propose an improved face feature extraction method based on two-dimensional Principal Component Analysis (2DPCA). This method obtain the feature space A based on overall covariance matrix and obtain the feature space B based on class covariance matrix. Finally, integrate the two feature spaces by weighting to form a new feature space for next classification. The class covariance matrix full use the information among the different types, so compared with 2DPCA, the improved algorithm has a better recognition rate.(3) For existed SVM may structure some non-important classification surfaces in training, develop a new improved SVM. The new algorithm decrease the training and recognition time by reducing the number of separating hyperplane.(4) Introduce the design and implementation process of face recognition system from the technical point of face recognition, including: system design concepts, technical processes, functional requirements analysis and system tools and platforms. Finally, the proposed improved 2DPCA is done simulation on this system. Compared with PCA, 2DPCA, LDA, LPP, etc., verify the proposed algorithm is superior to other algorithms in recognition rate.