中文摘要：

研究并实现了一种基于卷积神经网络的人脸识别方法.该网络由2个卷积层、2个池化层、1个全连接层和1个Softmax回归层组成,它能自动提取人脸特征并进行分类.网络通过批量梯度下降法训练特征提取器和分类器,各隐层应用＂dropout＂方法解决了过拟合问题.应用于ORL和AR人脸数据库的人脸识别率分别达到99.50%和99.62%,识别单张人脸的时间均小于0.05s,而且对光照差异、面部表情变化、有无遮挡物等干扰具有鲁棒性.

英文摘要：

Feature extraction and classification are two key steps in face recognition.A convolutional neural network composed of two convolutional layers,two pooling layers,one full-connection layer and one Softmax regression layer for face recognition is proposed.The neural network can automatically extract facial features and classify face,with trained feature extractors and the classifier using batch gradient descent.And the network adopts dropout method in hidden layers to avoid the overfitting problem.Experimental results show that proposed network achieves 99.50% recognition accuracy on ORL database and an accuracy of 99.62% on AR database,and it can complete one-time facial recognition in less than 0.05 s.More importantly,the network is robust to illumination variances,facial expressions and occlusions.