Research On Specific Human Face Recognition Technology Base On SVM

The specific human face recognition technology is an important part of human face recognition.Its primary purpose is to decide if one human face sample belongs to a particular class. From thealgorithm side, the specific human face recognition technology has many special requires indistribution of training samples, classify threshold, and the result valuation. First of all, the trainingsample in specific human face recognition often is small sample, and the threshold should setting forthe different class of samples. The evaluation of specific human face recognition can’t simplycalculate by accuracy of classify, and should judge by many target such as true accept rate, truereject rate, false accept rate, false reject rate.This paper had done some experiments of specific human face recognition by classicrecognition algorithm, point out the weakness though the performance of ROC curve: Beside thenature defect of these algorithms, the major reason of low performance in specific human facerecognition is that these algorithms can’t describe the wild distribution of negative samples.For increase the performance of classic algorithm in specific human face recognition, this paperdecides to combine the SVM classifier by AdaBoost algorithm to training the classifier. In thetraining process, the parameters of SVM classifier could be changed every turn to fit the class ofsamples, which can improve the classify accuracy of algorithm for different class of samples. In theother hand, because of the accuracy and variation requirement of weak classifier in AdaBoost, asteady classifier could hardly be improved by AdaBoost training. So, the variation must be insurewhen choose a weak classifier. Because of this, this paper use SVM classifier of RBF kernel function,and designed AdaBoost-OPSVM and AdaBoost-CPSVM to adjust the requirement in each trainingturn to improve the final recognition accuracy.