联合多模态表示的人脸识别方法研究  

Research on face recognition method based on multimodal representations

作　　者：张杜娟 陈飞 丁莉 ZHANG Dujuan;CHEN Fei;DING Li(School of Health Services Management,Xi’an MedicaI University,Xi’an 710021,China)

机构地区：[1]西安医学院卫生管理学院,陕西西安710021

出　　处：《电子设计工程》2021年第24期175-178,183,共5页Electronic Design Engineering

基　　金：陕西省教育厅专项科研计划项目(20JK0886);陕西省大学生创新创业训练计划项目(S202111840040);西安医学院大学生创新创业训练计划项目(121521040)。

摘　　要：针对人脸图像识别问题,提出联合多模态表示的方法。采用二维经验模态分解(BEMD)对人脸图像进行分解,获得多层次二维内蕴模函数(BIMF),作为人脸的多模态表示。在分类阶段,采用多任务压缩感知模型(MTCS)对多层次BIMF进行表征,求解不同BIMF的稀疏系数矢量,进而计算不同类别对于当前测试样本的重构误差,根据重构误差判断样本所属类别。通过在ORL和Yale-B人脸图像数据集上开展实验,所提方法平均识别率分别达到75.42%和94.07%,且对噪声干扰具有良好的稳健性。For face image recognition problem,a method using multimodal representations is proposed.The Bidimensional Empirical Mode Decomposition(BEMD)is applied to decompose the face image with multi⁃layer Bidimensional Intrinsic Mode Functions(BIMF),which are used as the multimodal representations for the face image.In the classification phase,the multi⁃task compressive sensing(MTCS)is adopted to represent the multi⁃layer BIMF to obtain their individual sparse coefficient vectors.Further,the reconstruction errors of different classes for the test image can be calculated.According to the reconstruction errors,the face label can be decided.Experiments are conducted on the ORL and Yale⁃B face image datasets,the proposed method could achieve average recognition rates of 75.42%and 94.07%,respectively,also with good noise robustness.

关 键 词：人脸识别 二维经验模态分解 多模态表示 多任务压缩感知 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]