Research Of Face Recognition Methods Based On Gabor Features And SVM

As intelligent technology development, based on face recognition automata application in every field is becoming more and more widely, and the face recognition system is to simulate the function of the human eye, through the acquisition of two-dimensional image of a face to face feature to implement automatic extraction and classification of the facial recognition.Face recognition technology is currently widely used biometric technology, through the extraction of facial features to achieve the identity of the automatic identification. Because of Gabor transform kernel function distribution and mammalian visual cortex simple cells in 2 d receptive field profiles are very similar, and it has good directional selectivity and spatial locality, to image local space scale of multiple directions in the area of information and local structural features of access provides a more effective method. In order to verify the effectiveness of Gabor features in face recognition, face recognition in this paper, the four traditional subspace method and the face recognition based on Gabor feature subspace method has carried on the comparative study, the use of ROC and CMC to verify the two parameters based on Gabor feature subspace face recognition method is effective and accurate. Verified experiments on ORL face database, the results show that under the same conditions, face recognition based on Gabor feature subspace method is of higher human face recognition rate and stronger robustness.On the basis of the study, on the basis of guarantee the recognition rate, further shorten the recognition time, Gabor features and SVM is proposed in this paper, based on optimization method of face recognition. First Gabor feature of all the training samples are extracted, and then put all the training samples extracted Gabor feature and the corresponding class label use Boosting algorithm to eliminate the non discriminant features for classification is meaningless, with remaining finally optimize the Gabor features to train the SVM classifier. The experimental results show that the optimized Gabor feature and the SVM classifier can improve the accuracy of face recognition, more important is the method of running time is short, provides the basis for the application promotion.