字典学习优化判别性降维的鲁棒人脸识别

Discriminative Dimensionality Reduction Optimized by Dictionary Learning for Robust Face Recognition

针对现有的人脸识别算法由于光照、表情、姿态、面部遮挡等变化而严重影响识别性能的问题,提出了基于字典学习优化判别性降维的鲁棒人脸识别算法。首先,利用经典的特征提取算法PCA初始化降维投影矩阵;然后,计算字典和系数,通过联合降维与字典学习使得投影矩阵和字典更好地相互拟合;最后,利用迭代算法输出字典和投影矩阵,并利用经l2-范数正则化的分类器完成人脸的识别。在扩展YaleB、AR及一个户外人脸数据库上的实验验证了本文算法的有效性及鲁棒性,实验结果表明,相比几种线性表示算法,本文算法在处理鲁棒人脸识别时取得了更高的识别率。

The performance of face recognition system is seriously impacted by illumination, expression, pose and occlusion variations, for which the algorithm of discriminative dimensionality reduction optimized by dictionary learning is proposed. Firstly, typical feature extraction algorithm PCA is used to initialize dimensionality reduction projection matrix. Then, dictionary and coefficient is computed and the dictionary can match with each other by jointing dimension reduction and dictionary learning. Finally, dictionary and projection matrix is outputted by using iterative algorithm, and classifier regularized by l2-norm is used to finish face recognition. The effectiveness and reliability of proposed algorithm has been verified by experiments on extended YaleB, AR and a wild face databases. Experimental results show that proposed algorithm has higher recognition accuracy than several other linear represent algorithms when dealing with robust face recognition.