基于半监督学习的最大间距准则人脸识别

Maximum Margin Criterion Face Recognition Based on Semi-supervised Learning

针对人脸识别中不能同时利用标记样本和无标记样本的问题,提出了一种具有半监督学习的最大间距准则人脸特征提取方法.算法把无监督判别投影的局部和非局部散度矩阵引入到最大间距准则方法中.在ORL和Yale人脸库上的实验结果表明,本方法不但充分利用了有判别信息的标记样本和大量无标记样本,而且更好地发挥了最大间距准则算法的优点,有效地提高了人脸识别率.

In view of the problem that the labeled samples and unlabeled samples cannot be used at the same time in face recognition, a new face feature extraction method which is characterized by semi-super- vised learning and maximum margin criterion is proposed. The algorithm introduces the un-supervised dis- criminant projection of local and nonlocal scatter matrix into the maximum margin criterion method. The experimental results from ORL and Yale face database show that the method not only makes full use of the labeled samples and a large number of unlabeled samples with discriminant information, but also brings out the advantages of the maximum margin criterion. Therefore, the face recognition rate is efficiently enhanced.