基于SIFT特征的单训练样本人脸识别

针对RoboCup竞赛家庭组比赛对人脸识别的要求,研究了单训练样本的人脸识别问题。设计了一种基于尺度不变特征变换的单训练样本人脸识别系统。将人脸区域划分为4×4的均匀网格,然后在每个区域选取最具有判别能力的少量特征,再将这些特征与测试人脸对应网格的特征进行匹配;采用加权和形式评价测试人脸和每个训练人脸的匹配程度,选择阈值进行识别决策。使用ORL人脸数据库对设计的识别系统进行了测试,结果表明,设计的系统可以达到100%的识别精度和70%的检出率,可以达到比赛的要求。

Face recognition from a single training image-a sift approach

Face recognition from a single training image is studied according to the desired ability of face recognition in Home Leagues of RoboCup.A face recognition system from a single training image is designed by using scale invariant feature transform.First,the face image is divided into 4×4 uniform grids.Second,a few discriminative features are selected within every region.Then those features are matched with those in the corresponding grid of test image.The final matching score is the weighted sum of every matching score between the features in the corresponding grid of test image and training image.The decision is made by a given threshold.The proposed face recognition system is tested using ORL face databases.The results show that 100% recognition precision and 70% recall performance are realized by the designed system,which can meet the demand of @Home Leagues of RoboCup.