基于小波变换的多字典人脸识别方法  

Multi dictionary face recognition method based on wavelet transform

作　　者：李舜 姚明海[1] 李劲松 王娜 LI Shun;YAO Minghai;LI Jinsong;WANG Na(College of Information Science and Technology,Bohai University,Jinzhou 121013,China)

机构地区：[1]渤海大学信息科学与技术学院,辽宁锦州121013

出　　处：《渤海大学学报（自然科学版）》2022年第2期185-192,共8页Journal of Bohai University:Natural Science Edition

基　　金：辽宁省自然科学基金项目(No:2019-ZD-0503);辽宁省教育厅科学技术研究项目(No:LJ2020003,No:WJ2020004).

摘　　要：传统基于稀疏表示的人脸识别方法因未充分利用样本包含的信息,而存在较低的鲁棒性和识别率等问题.提出基于小波变换的多字典人脸识别方法.提取人脸图像的多尺度纹理特征,构建多字典,每个字典分别对应样本的原始图像和不同尺度的纹理信息.为提高人脸识别的准确性,在训练阶段利用相同的系数表示将不同字典相关联,深入挖掘了人脸图像与其不同尺度纹理间的共性,探索人脸不同特征的内在联系.较仅聚焦于样本单一特征的传统方法,克服了忽视样本不同特征间联系的不足,更深刻地挖掘训练样本人脸特征且突出了不同类训练样本的个性特征.在多个人脸数据库上的实验结果表明,识别性能有明显的提升.The traditional face recognition method based on sparse representation has some problems,such as low robustness and recognition rate,because it does not make full use of the information contained in the sample.A multi dictionary face recognition method based on wavelet transform is proposed.The multi-scale texture features of face images are extracted and multi dictionaries are constructed.Each dictionary corresponds to the original image and texture information of different scales.In order to improve the accuracy of face recognition,the same coefficient representation is used to associate different dictionaries in the training stage,the commonness between face images and their different scale textures is deeply excavated,and the internal relationship between different features of face is explored.Compared with the traditional method that only focuses on a single feature of the sample,it overcomes the deficiency of ignoring the relationship between different features of the sample,excavates the face features of the training sample more deeply,and highlights the personality characteristics of different types of training samples.The experimental results on multiple face databases show that the recognition performance has been significantly improved.

关 键 词：字典学习 稀疏表示 人脸识别 小波变换 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]