基于支持向量机和Candide-3的单样本人脸确认  

FACE VERIFICATION BASED ON SVM AND CANDIDE-3 WITH SINGLE SAMPLE

作　　者：马俊容[1] 胡峰松[1] 林斌[1] 

机构地区：[1]湖南大学计算机与通信学院,湖南长沙410082

出　　处：《计算机应用与软件》2010年第6期213-215,共3页Computer Applications and Software

摘　　要：目前已有的正面人脸识别方法大多要求有充分数量的训练样本,不能适应单人单个训练样本的情况。提出了一种基于支持向量机SVM的单训练样本人脸确认方法,首先产生人脸Candide-3模型,通过对重建模型的旋转产生姿态不同的数字人脸,将产生的数字人脸和原始样本一起作为训练数据,用SVM进行分类确认,取得了接近于多样本情况下的识别率。在ORL人脸库的实验结果表明,本方法能更好地适应单样本情况下的人脸确认,比同类人脸确认方法具有更高的识别率。Most of existing front face recognition algorithms require sufficient samples for training but do not adapt to the cases of single sample per person.In this paper,we proposed a useful method of face verification,which is based on Support Vector Machines（SVM） with single training sample.First we formed a human face＇s Candide-3 model,and then by rotating the reconstructed model we made digital face with different attitudes.These digital faces obtained were taken as the training data together with the primary samples and then used SVM to classify and verify them,the achieved recognition rate verged on that in the case of multiple sample.Experiments on the ORL facebase shown that the method we proposed can better adapt to face recognition in single sample condition;it has a higher recognition rate in the face verification methods of its kind.

关 键 词：人脸确认 单训练样本 支持向量机 Candide-3 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]