基于判别字典学习的线性子空间人脸识别算法  

Linear subspace face recognition algorithm based on discriminant dictionary learning

作　　者：任克强[1] 张静然 REN Keqiang;ZHANG Jingran(School of Information Engineering,Jiangxi University of Science and Technology,Ganzhou 341000,Jiangxi China)

机构地区：[1]江西理工大学信息工程学院

出　　处：《华中科技大学学报（自然科学版）》2019年第9期95-100,共6页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61501210)

摘　　要：提出判别字典学习来获取线性子空间方法,以减弱光照等噪声对子空间人脸特征提取的影响,从而在保证稀疏系数的局部结构性同时保持字典的判别性.首先,训练与语意相关的结构字典,并在破坏非同类语意样本间局部结构稀疏性的同时,增强同类语意样本间局部结构的稀疏性;其次,利用最大间隔准则(MMC)在重构后稀疏易分的语意子空间对样本进行特征提取,不仅可以避免小样本问题还可以在重构后的语意空间中提取抗噪声干扰的特征.在Yale库、AR库和Yale B库数据集上的试验结果表明:与现有算法相比,该算法有更优的性能,能更高效地提取不受噪声干扰的易分类人脸语意特征.A discriminative dictionary learning method was proposed to obtain the linear subspace method to reduce the influence of illumination and other noise on the subspace facial feature extraction.To ensure the discriminability of the dictionary while maintaining the local structure of the sparse coefficient,firstly,the syntactic-related structural dictionary was trained.And while destroying the sparseness of local structure between non-similar semantic samples,the sparseness of local structure between similar semantic samples was enhanced.Secondly,the maximum interval criterion(MMC)was used to characterize the samples in the sparse and easy-separated semantic subspace after reconstruction.The extraction can not only avoid the small sample problem but also extract the anti-noise interference feature in the reconstructed semantic space.The experimental results on the Yale library,the AR library and the Yale B library data set show that compared with the existing algorithm,the algorithm has better performance and can extract the semantic features of easy-to-classify faces without noise interference.

关 键 词：线性子空间 人脸识别 稀疏表示 字典学习 判别投影 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP181[自动化与计算机技术—计算机科学与技术]