The Method Research Of The Face Image Based On Neural Network

In the world，Face Recognition has always been a hot research topic. It hasapplication value of extensive and potential. In the study of the face recognition, how toextract the useful feature effectively and classifying are the difficult problem.Meanwhile, the recognition rate has been the key problem in the design of the facerecognition system.This paper is based on the methods of subspace feature extraction and the NationalNatural Science Foundation project on” The research of the artificial brain, Based on theconnected character homologue, A pattern recognition the new neural network model”No.61272077, and the funds of the scientific research of Nanchang Hangkonguniversity “Based on the connected character homologue. A pattern recognition of theartificial brain the new neural network model”, No.EA2009003, to research the patternrecognition neural network model and the methods of face recognition. In this paper, theresearch contents and innovation points as follow：1Implements the located function of the face, eye, nose and mouth base on humanface skin color, by designed the software.2In order to solve the problem of the feature extracted is not enough, the maincharacteristics can’t be extracted and the dimension too high of the characteristics, In thethesis, proposed the method of extract feature by principal component projection of theclass distance of within and between.3In order to solve the problem of the speed of clustering center convergence can’tbe adjusted, in the thesis, proposed the method of the improved RBF neural network,that is, the neural network of variable clustering center.4Finally, in the thesis, proposed the method of the design method of the facerecognition system, the effectiveness of the proposed feature extraction method andclassification method was verified by the experiment on ORL face database. At sametime, the result suggests that some aspects need to be improved.