Study Of The Application Of Principal Component Analysis To Face Recognition

Face recognition is a hot subject in the current research on the technology in the field of identity authentication. It involves the world of pattern recognition, image processing, neural network, computer vision, physiology and cognitive science, etc. The research result of face recognition has great importance of theoretical meaning and practical value.Firstly, the discussion to image preprocessing and classifier designing which are parts of face recognition system is carried out and a brief introduction of several feature exaction methods which are based on the global is given in this paper. Then, several feature exaction algorithms called PCA, 2DPCA, DiagPCA, (2D)~2PCA which are based on principal component analysis are described in detail. The algorithms of 2DPCA, DiagPCA, (2D)~2PCA are directly based on image matrices, having a relatively small computational complexity. As a result, those algorithms have a higher feature exaction rate than that of PCA algorithm. Meanwhile, during analyzing the recognition performance of the algorithms mentioned in this paper, it is found that the recognition speed of the algorithm is relatively quick under the condition that representative facial images are selected as training set samples, when the number of the training samples of each category is determined.Based on analyzing and comparing the feature exaction algorithms mentioned above, a new face recognition method called average face plus (2D)~2PCA is proposed in this paper. It is proved by the experiments that the new method needs small amount of comparisons among image features when the features are classified and has a fast recognition speed.Finally, the application of neural network in the face recognition method based on two dimension principal component analysis is preliminarily researched. And the method of the combination of (2D)~2PCA feature and BP neural network classifier is proposed. The new method is proved good performance by comparing it with the other methods introduced in the paper.