基于 Gabor 小波和 LBPH 的实时人脸识别系统-《计算机技术与发展》

文章信息/Info

作者:: 张伟 ¹ ; 程刚 ² ; 何刚¹ ; 阎石¹; 1. 兰州大学信息科学与工程学院,甘肃兰州 730000;2. 兰州大学网络安全与信息化办公室,甘肃兰州 730000

Author(s):: ZHANG Wei¹ ; CHENG Gang² ; HE Gang¹ ; YAN Shi¹; 1. School of Information Science and Engineering,Lanzhou University,Lanzhou 730000,China;2. Office of Network Security and Information,Lanzhou University,Lanzhou 730000,China

摘要:: 针对当前流行的人脸识别算法存在的光照敏感、鲁棒性差等局限性,为提高识别效率,提出了一种基于 Gabor 小波和 LBPH 算法的实时人脸识别算法。首先将人脸图像与 Gabor 核函数进行卷积得到人脸特征图像,并将其串联在人脸特征空间;然后使用 LBPH 提取该空间人脸特征,并通过匹配 LBPH 直方图序列的相似度来实现分类。此外,基于 ARM 平台对文中算法进行实现,形成了实时人脸识别系统。由于系统使用 MySQL 数据库管理人脸数据,因此可以实现实时的管理效果。该系统在自建单样本人脸数据库中能达到 92%的识别率,相比传统算法识别率更高、实时性更强。

Abstract:: Aiming at the limitations of illumination sensitivity and poor robustness for current popular face recognition algorithm,in orderto improve the recognition efficiency,we propose a real-time face recognition algorithm based on Gabor wavelet and LBPH algorithm.We first fuse the face feature image obtained by convolution of the face image with the Gabor kernel and concatenate it in the face feature space,and then use LBPH to extract the facial feature of the space and achieve the classification by matching the similarity of the LBPH histogram sequence. Moreover,a real-time face recognition system is established based on ARM platform. As the system uses theMySQL database to manage face data,dynamic management can be achieved. The system achieves 92% recognition rate in self built single face database. Compared with traditional algorithm,it has higher recognition rate and stronger real-time performance.