| Human face recognition is widely used in many areas such as enterprises secrecy, management of social population, criminal investigation, judicial proof, national security and global anti-terrorism etc. For decades, many students refer to the mathematic research results about geometric feature analysis, statistical feature analysis, kernel function analysis, and neural networks. Study and explore about effectively feature extraction and recognition methods. Human face recognition has been the hotspot of image process.Images preprocessing, feature extraction and image recognition are presented in this paper, which resolve the contradiction between identification speed and its rate.Several methods of images preprocessing are given in the paper: image denoising, image enhancement, size normalization, light compensation and image division. Through these transformations , image can reduce the disturbance factor as least as possible,also may intensify the obviously characteristic and weaken the non—characteristic part of image.The feature of human facial and its small-area images are fused in this paper. An extended optimal Gabor feature sampling algorithm is proposed to reduce the dimension of human facial features. Then extract features by principal component analysis. The method improves the speed of feature extraction and its reliability.Deeply researches on SVM in solving small sample problems and the high efficiency of binary-tree. The decision method of multi-classification is based on the combination of SVM and binary-tree. Development of a characters selection method of SVM based on Particle Swarm Optimizer, which effectively accelerates the characters selection. Simulate and experiment on the methods of images preprocessing, feature extraction and face recognition respectively from the standard database of ORL. Experiment results show that the method presented has good characteristic such as high efficiency, good reliability, and strong practicability. |
PDF Full Text Request