Technology Research Of Face Recognition Based On PCA

Face recognition is an active research field in identification technology of living things and the intelligent software of face recognition is significant to the new period such as anti-terrorism and synthesizing safety. Face recognition system uses human images as its recognition object .It adopts the technology of computer vision and image processing to find the contour of human faces and part of detail face features, then studies the recognition methods using the found features .At present, face recognition continues to be a hot topic in pattern recognition field due to its wide range of applications such as distinguishing somebody from others and controlling authority. Many face recognition methods have been proposed and they are typically divided into two. One is based on the integral part of the facial features; the other is based on the part of facial features. This thesis investigates the use of the engine faces method (PCA) based on the integral part of the facial features and recognition. The main work reads as follows:(1)The background of the research and the latest development of face recognition methods are introduced.(2)The main parts of the face recognition are face detection and feature extraction, so in this thesis we first realize a system of face detection and fixer. The system detects the whole face using the method of complexion model, fatherly it finds the positions of eyes and mouth and then an accurate face is found. In the test we discover this system can fast detect faces upright or rotated with small angle in color images.(3)First, the thesis investigates principle component analysis (PCA) approach deeply, and then the choice of feature vector of sample's covariance matrix and distance measure criterion are discussed. PCA approach can approximate the original data with lower dimension feature vector. Just because of this, it takes more time on feature extraction .But 2DPCA can take less time on calculation. Here the two methods are compared on theory and experimental data, and it's showed that 2DPCA is superior to the traditional PCA. (4)The methods based on fisher discriminant analysis for face recognition are investigated. The traditional LDA method often suffers the small sample size problem, so many new methods are proposed to solve this shortcoming, and PCA + LDA is one of them. This method obtained the feature sub-space of training samples by way of PCA, and feature sub-space from LDA is also calculated on the PCA. Then the two feature sub-spaces are fused, and the fusion feature space is acquired. Recognition features are gained after training samples and test samples are respectively projected towards the fusion feature space. Then a new method is proposed-2DPCA+LDA.The experimental on ORL face database shows that 2DPCA+LDA is better than traditional PCA+LDA and 2DPCA.