通用学习框架结合2DPCA的鲁棒人脸识别

针对现有的人脸识别算法由于光照、表情、姿态、伪装等变化而严重影响识别性能的问题,提出了一种基于通用学习框架结合2DPCA的鲁棒人脸识别算法。首先借助于额外的通用训练样本集进行样本的叠加以增加训练样本的数量;然后利用经典的2DPCA算法进行特征提取;最后,利用最近邻分类器对人脸进行分类并完成最终的人脸识别。在基准人脸数据库ORL、FERET及鲁棒人脸数据库AR、扩展YaleB上的实验验证了该算法的有效性及鲁棒性,实验结果表明,相比其他几种人脸识别算法,提出的算法不仅提高了人脸识别率,而且大大地减少了识别所用时间,有望应用于实时鲁棒人脸自动识别系统中。

Robust Face Recognition Based on Fusion of 2DPCA with Generic Learning Framework

The recognition performance of existing algorithms is seriously impacted by variation of illustration, expression, pose and mask, for which a face recognition algorithm based on 2DPCA improved by generic learning framework is proposed. Firstly, Training samples are composited with additional generic learning training samples to increase the number of training samples. Then, classical 2DPCA is used to extract features. Finally, nearest neighbor classifier is used to classify and finish the face recognition work. The effectiveness and robustness of proposed algorithm has been verified by experiments on the two baseline face database ORL, FERET and robust face databases AR and extended YaleB. Experimental results show that proposed method has higher recognition accuracy and less time taken comparing with several advanced algorithms, which indicates that it is expected to be applied into robust real-time face recognition system.