The Research On Face Recognition Methods Based On Support Vector Machines

Automated face recognition technique is one of the most attractive branches ofbiometrics and it is also one of the most active and challenging tasks for computervision and pattern recognition in recent30years. For the best security, reliability andvalidity, it is not only widely applied in a variety of personal identification systemsuch as national security, public security, justice, government, finance, business andalso can be used in the field of human-computer interface and visual communication.This dissertation mainly studies the approaches to frontal face recognition ingray images. The main contents are as follows:(1) Make use of PCA to extract the feature information of images.Through thismethod to extract facial features of vector images and principal components, in orderto be a high-dimensional mapping of the human face image to the characteristics oflow-dimensional space (features of the face), making the simplified facial imagewhile retaining most of its information. So will the PCA algorithm feature extractionand detection of data as a feature vector for Face Recognition(2) Based on the "similar" SVM methods of face recognition.The classifier isused so called "similar "to construct a number of the second SVM category, thismethods make face recognition as a typical problem of the many types ofdiscrimination into the second category. Unlike the previous One-against-One andOne-against-Rest method, the method only for the N category structure N request, arequest through the N output value to determine a given face with certain types ofdata in the database Similar size, which will be to determine the correct face of thecategory. Since this method than in the past methods classifier greatly reducing thenumber, it guarantees the accuracy of face recognition on the basis of calculation toreduce the complexity, reduce the computing time.(3) Face Recognition System Design based on SVM. In order to improverecognition accuracy rate, added a neural network training device. The recognitionexperiment result show a good performance in ORL face database.