基于改进的减法平均优化算法与BP神经网络的人脸识别  

Face Recognition Based on Improved Subtraction Average‑Based Optimizer and BP Neural Network

作　　者：杨泽锐 胡红萍[1] YANG Zerui;HU Hongping(School of Mathematics,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学数学学院,山西太原030051

出　　处：《中北大学学报（自然科学版）》2024年第6期725-736,共12页Journal of North University of China(Natural Science Edition)

基　　金：山西省基础研究计划资助项目(20210302123019,20210302124195,20210302124212,20210302123189);山西省回国留学人员科研资助项目(2020-104,2021-108)。

摘　　要：本文在减法平均优化算法(Subtraction Average-Based Optimizer,SABO)的初始化阶段引入了混沌映射,并与黄金正弦算法结合,提出了改进的减法平均优化算法(Improved Subtraction Average-Based Opti-mizer,ISABO),解决了减法平均优化算法可能陷入到局部最优解的问题,并通过23个基准函数的极值寻优验证了ISABO的有效性。针对静态的人脸图像的分类识别问题,本文依次利用直方图均衡化处理方法和高斯滤波处理方法进行图像预处理,再利用主成分分析法(Principal Component Analysis,PCA)对图像进行特征提取,最后利用ISABO算法优化BP神经网络实现人脸图像分类,这样建立了基于ISABO和BP神经网络的人脸识别模型ISABO-BP。实验结果表明,本文提出的ISABO-BP在ORL人脸数据库的人脸识别平均准确率为97.50%,优于其他比较算法,并且拥有良好的稳定性,有效降低了误识率、拒识率以及拒错比。Chaos mapping was introduced in the initialization stage of subtraction average-based optimizer(SABO)and combined with golden sine algorithm.The improved subtraction average-based optimizer(ISABO)was proposed to solve the problem that the subtraction average-based optimizer might fall into the local optimal solution,and the effectiveness of ISABO was verified by the extremum optimization of 23 reference functions.Aiming at the problem of classification and recognition of static face images,this paper used histogram equalization processing method and Gaussian filter processing method successively for image preprocessing,and then used principal component analysis(PCA)to extract image features.Finally,the face recognition model ISABO-BP based on ISABO and BP neural network was established by optimizing BP neural network to realize face image classification.The experimental results show that the face recognition accuracy of the proposed ISABO-BP in ORL face database is 97.50% on average,which is better than other comparison algorithms,and has good stability,effectively reducing the error rate,rejection rate and error rejection ratio.

关 键 词：人脸识别 主成分分析法 减法平均优化算法 黄金正弦算法 混沌映射 BP神经网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]