基于扩展字典学习的鲁棒人脸识别算法

Robust face recognition algorithm based on extended dictionary learning

针对传统人脸识别的有限训练样本字典学习不稳定、噪声处理不够鲁棒、运行速度慢等问题,提出了一种扩展字典学习的鲁棒人脸识别算法。首先利用原始训练样本生成两个扩展训练样本,然后对两个扩展训练样本添加以像素损坏为主要影响因素的随机噪声和以遮挡为主要影响因素的结构噪声,通过增加训练样本的多样性,获得更鲁棒的字典。实验结果表明,该算法在Extended Yale B、AR、ORL数据库上识别率高,对噪声鲁棒性强,运行速度快。

Traditional face recognition is not robust in noise processing,slow in running speed and unstable in dictionary learning.This paper proposes a robust face recognition algorithm based on extended dictionary learning.Firstly,two extended training samples were generated from the original training samples.Then,the random noise with pixel damage as the main influencing factor and the structural noise with occlusion as the main influencing factor were added to the two extended training samples.By increasing the diversity of training samples,a more robust dictionary was obtained.The results show that the algorithm has high recognition rate,strong robustness to noise and fast running speed on Extended Yale B,AR and ORL databases.