Abstract: At present, the outstanding problems encountered in face recognition technology are the decline of recognition precision caused by illumination, attitude, occlusion and expression. These problems are the main reasons for the imperfection of face recognition system. Deep learning is a new way to solve these problems effectively. Firstly, the deep learning algorithm is introduced to the multi-level learning, and then the high-level features are extracted to describe the human face. Finally, the Maximum Margin Criterion is used to reduce the reconstruction error caused by the least square estimation and realize the effective face recognition classification. The algorithm is simulated in the ORL, CAS-PEAL and the extended Yale-B face database under different illumination, attitude, occlusion, expression and changes in facial features conditions. The results show that the proposed algorithm has higher efficiency and accuracy than the traditional linear classification algorithm.

Key words: face recognition, deep learning, Maximum Margin Criterion（MMC）, the least square estimation

摘要： 当前，人脸识别技术遇到的突出问题是光照、姿态、遮挡和表情等因素所引起的识别精度的下降，这些问题是人脸识别系统不完美的主要原因，深度学习是一种新的方法，可有效解决这些问题。首先通过引入深度学习算法进行多层次的学习，然后提取高层特征进行人脸描述，最后应用最大间距准则减小最小二乘估计产生的重建误差，实现有效的面部识别分类。该算法在ORL、CAS-PEAL和扩展Yale-B人脸数据库中进行了不同光照、姿态、遮挡、表情和容貌特征变化条件下的仿真实验。结果表明，所提出的算法比传统线性分类算法具有更高的效率和准确度。

关键词: 人脸识别, 深度学习, 最大间距准则, 最小二乘估计