基于改进PCA+SVM的人脸识别系统

doi:10.16180/j.cnki.issn1007-7820.2021.12.010

摘要/Abstract

摘要：

PCA算法对数字图像处理人脸识别率不高,且无法处理人脸的非线性特征。针对这一问题,文中在原有PCA算法的基础上提出了基于KPCA并结合SVM的人脸识别研究方法。该方法通过利用核改进PCA后的KPCA算法的内部非线性核函数提取人脸面部轮廓,处理非线性特征数据并降低数据维数,有效减少特征数据存储所需的空间并提高了运算能力。然后,再结合SVM分类器进行分类识别,可较好地提高系统识别率。实验结果表明,文中所提算法在ORL人脸库中的识别率为95.16%,在Yale人脸库中识别率为95.10%。在MATLAB平台上搭建的系统能正确识别出人脸,证明了文中所提系统的可行性,为实际研究提供了参考。

关键词: 人脸识别, 主成分分析, 核主成分分析, Gamma校正法, 支持向量机, 降维, 分类识别, GUI

Abstract:

The PCA algorithm has low recognition rate for digital image processing, and cannot deal with the non-linear features of the face. In view of this problem, in the basis of original PCA algorithm, a face recognition research method based on KPCA combined with SVM is proposed in this study. By using the internal non-linear kernel function of the KPCA algorithm after the PCA is improved by the kernel, the facial contours of the face are extracted, the non-linear feature data is processed and the data dimension is reduced, which can better reduce the space required for feature data storage and improve the computing power. Then, combined with the SVM classifier for classification and recognition, the system recognition rate is improved. Experiments show that the recognition rate of the proposed algorithm in the ORL face database is 95.16%, and the recognition rate in the Yale face database is 95.10%. The system established on MATLAB can correctly recognize human faces, which proves the feasibility of the system proposed in the study, and has certain reference value for actual research.

Key words: face recognition, principal component analysis, KPCA, Gamma correction method, SVM, dimensionality reduction, classification recognition, GUI

中图分类号:

TP391.4

彭荣杰,彭亚雄,陆安江. 基于改进PCA+SVM的人脸识别系统[J]. 电子科技, 2021, 34(12): 56-61.

PENG Rongjie,PENG Yaxiong,LU Anjiang. Face Recognition System Based on Improved PCA+SVM[J]. Electronic Science and Technology, 2021, 34(12): 56-61.

图/表 7

图1

图2

图3

图4

表1

表2

图5

参考文献 19

[1]	李梦潇, 姚仕元. 基于PCA的人脸识别系统的设计与改进[J]. 计算机科学, 2019, 46(S1):577-579.
	Li Mengxiao, Yao Shiyuan. Design and improvement of face recognition system based on PCA[J]. Computer Science, 2019, 46(S1):577-579.
[2]	易焱, 蒋加伏. 基于LBP和栈式自动编码器的人脸识别算法研究[J]. 计算机工程与应用, 2018, 54(2):163-167.
	Yi Yan, Jiang Jiafu. Face recognition based on LBP and stacked autoencoder[J]. Computer Engineering and Applications, 2018, 54(2):163-167.
[3]	韩春雷, 蔚婧. 近邻数据分类器在多目标跟踪中的应用研究[J]. 电子科技, 2018, 31(6):32-35.
	Han Chunlei, Yu Jing. Implementation of nearest neighbor data classifier in multi-target tracking[J]. Electronic Science and Technology, 2018, 31(6):32-35.
[4]	邹鼎杰. 基于知识图谱和贝叶斯分类器的图书分类[J]. 计算机工程与设计, 2020, 41(6):1796-1801.
	Zou Dingjie. Book classification based on knowledge graph and Bayesian classifier[J]. Computer Engineering and Design, 2020, 41(6):1796-1801.
[5]	李昆仑, 张炘. 级联SVM和分类器融合的人脸性别识别方法[J]. 计算机工程与应用, 2017, 53(8):154-158.
	Li Kunlun, Zhang Xin. Hierarchy support vector machines and classifier fusion methods for face image gender recognition[J]. Computer Engineering and Applications, 2017, 53(8):154-158.
[6]	郭晨晨, 朱红康. 基于粗糙集理论的模糊支持向量聚类算法的改进[J]. 洛阳理工学院学报(自然科学版), 2016, 26(3):69-73.
	Guo Chenchen, Zhu Hongkang. Improvement fuzzy SVC algorithm based on rough set theory[J]. Journal of Luoyang Institute of Science and Technology(Natural Science Edition), 2016, 26(3):69-73.
[7]	金玉多. 改进Gamma校正的人脸图像光照预处理算法[J]. 脑知识与技术, 2019, 15(20):214-217.
	Jin Yuduo. Preprocessing algorithm of face illumination with improved Gamma correction[J]. Computer Knowledge and Technology, 2019, 15(20):214-217.
[8]	邓金. 基于自然语言处理技术的邮件检测系统[J]. 数字技术与应用, 2019, 37(6):117-118.
	Deng Jin. Mail detection system based on natural language processing technology[J]. Digital Technology & Application, 2019, 37(6):117-118.
[9]	韩鹏, 张代远. 一种基于光照方向估计的人脸识别方法[J]. 计算机技术与发展, 2012, 22(6):85-88.
	Han Peng, Zhang Daiyuan. Face recognition based on illuminant direction estimation[J]. Computer Technology and Development, 2012, 22(6):85-88.
[10]	Hu L, Cui J. Digital image recognition based on Fractional-order-PCA-SVM coupling algorithm[J]. Measurement, 2019, 14(5):150-159.
[11]	姚立平, 潘中良. 一种多算法融合的人脸识别方法研究[J]. 光电子·激光, 2019, 30(9):960-967.
	Yao Liping, Pan Zhongliang. Research on a face recognition method based on multi-algorithms fusion[J]. Journal of Optoelectronics·Laser, 2019, 30(9):960-967.
[12]	穆新亮. 基于混合核函数的快速 KPCA 人脸识别算法[J]. 电子科技, 2015, 28(2):46-50.
	Mu Xinliang. A fast KPCA face recognition algorithm based on mixed kernel function[J]. Electronic Science and Technology, 2015, 28(2):46-50.
[13]	Kempfert K C, Wang Y S, Chen C X, et al. A comparison study on nonlinear dimension reduction methods with kernel variations: Visualization, optimization and classification[J]. Intelligent Data Analysis, 2020, 24(2):267-290. doi: 10.3233/IDA-194486
[14]	刘芬. 基于线性和非线性算法的高效人脸识别系统[J]. 电子科技, 2019, 32(7):82-86.
	Liu Fen. An Efficient face recognition system based on linear and nonlinear algorithms[J]. Electronic Science and Technology, 2019, 32(7):82-86.
[15]	蒋先刚, 张盼盼, 胡传秀, 等. 视频烟雾的颜色和动态特征的选择及探测方法[J]. 计算机工程与设计, 2016, 37(7):1867-1872.
	Jiang Xiangang, Zhang Panpan, Hu Chuanxiu, et al. Feature selection and smoke detection method based on color and motion in video[J]. Computer Engineering and Design, 2016, 37(7):1867-1872.
[16]	张珂, 宋文丽, 石怀涛, 等. 基于改进核主元分析的故障检测方法研究[J]. 控制工程, 2017, 24(2):418-424.
	Zhang Ke, Song Wenli, Shi Huaitao, et al. Fault detection based on improved kernel principal component analysis[J]. Control Engineering of China, 2017, 24(2):418-424.
[17]	Dong J H, Gu J, Ma X, et al. Face recognition based on KPCA and SVM [C].Shenyang:Proceeding of the Eleventh World Congress on Intelligent Control and Automation, 2014.
[18]	黄华盛, 杨阿庆. 基于PCA 算法的人脸识别[J]. 电子科技, 2015, 28(8):98-101.
	Huang Huasheng, Yang Aqing. Face recognition based on PCA algorithm[J]. Electronic Science and Technology, 2015, 28(8):98-101.
[19]	杨涵. 基于深度学习的人脸识别算法研究[J]. 电子设计工程, 2019, 27(19):155-159.
	Yang Han. Research on face recognition algorithm based on deep learning[J]. Electronic Design Engineering, 2019, 27(19):155-159.

算法	训练样本数	测试样本数	识别率/%	平均运行时间/s
PCA	5	5	77.00	4.45
PCA+SVM	5	5	84.00	4.38
KPCA+SVM	5	5	95.16	3.95

算法	训练样本数	测试样本数	识别率/%	平均运行时间/s
PCA	5	5	74.00	5.27
PCA+SVM	5	5	83.10	6.22
KPCA+SVM	5	5	95.10	5.12