Face Recognition Based On Feature Fusion

Face recognition has drawn considerable interest and attention from many researchers and has wide range of potential applications in both public life and security. Currently, the outstanding face recognition system can achieve satisfactory performance under ideal conditions. However, face recognition performance needs to be improved under non-ideal conditions(main factors: illumination, pose, expression and occlusion). So, we have to solve many problems to establish a robust, practical face recognition system.Analysis the characteristics of the human face of global and local features, based on complementarity principle, combining global and local features by neural network to design classifier in order to improve the accuracy of face recognition. As a whole, the main content of this paper as follows:1. Recognize faces using global features of human face and determine the method of global feature extraction. Use the method of principal component analysis(PCA) to extract global features of face, compared with method of 2D-PCA through experiment. Determine to obtain the final face global features by the way of PCA.2. Use the method of Gabor transform to extract global features of face, use the tactics of fusion of multiple classifier to face recognition. Gabor-based face recognition method which is robust to the variations of expression and occlusion, high dimension of Gabor features makes it hard to the classifier to classify. Use the method of Gabor features packet and fusion to face representation and face recognition based on key parts of the human face. Integrate multiple features by assigning different weights to different parts to simplify the high-dimensional of Gabor feature.3 Fusion global and local features through Back Propagation(BP) neural network parallel. Design classifier to face recognition on this basis. Psychology research results show that the human visual perception system to recognize human faces at the same time using the global and local features, through the neural network integrate features. By teacher signal, to train the network, determine the weights to form a classifier. The results show that the neural network through global and local features fusion, can improve the accuracy of face recognition.