第14卷第2期 智能系统学报 Vol.14 No.2 2019年3月 CAAI Transactions on Intelligent Systems Mar.2019 D0:10.11992/tis.201709033 网络出版地址：http:/kns.cnki.net/kcms/detail/23.1538.TP.20180418.0946.006.html 图正则测化字典对学习的轻度认知功能障碍预测 魏彩锋2，孙永聪2，曾宪华2 (1.重庆邮电大学计算机科学与技术学院，重庆400065,2.重庆邮电大学计算智能重庆市重点实验室，重庆 400065) 摘要：针对字典对学习(DPL方法只考虑了同类子字典的重构误差和不同类表示系数的稀疏性，没有考虑图 像间的几何近邻拓扑关系的问题。通过近邻保持使得在同类近邻投影系数之间的距离较小，而不同类投影系 数之间的距离大，能够有效提高字典对学习算法的分类性能，基于此提出了基于几何近邻拓扑关系的图正则化 的字典对学习(GDPL)算法。在ADNII数据集上对轻度认知功能障碍预测的实验表明，使用GDPL算法学习 的编码系数作为特征预测的准确率(ACC)和ROC曲线下的面积(AUC)比使用结合生物标志作为特征预测的 准确率提高了2%6%.使用GDPL算法比DPL算法的实验结果也有提高。 关键词：图正则化：字典对学习；几何近邻关系：图像分类；轻度认知功能障碍预测 中图分类号：TP391:R749 文献标志码：A文章编号：1673-4785(2019)02-0369-09 中文引用格式：魏彩锋，孙永聪，曾宪华.图正则化字典对学习的轻度认知功能障碍预测.智能系统学报，2019,14(2)： 369-377. 英文引用格式：VEI Caifeng,SUN Yongcong,ZENG Xianhua..Dictionary pair learning with graph regularization for mild cognit- ive impairment prediction[J.CAAI transactions on intelligent systems,2019,14(2):369-377. Dictionary pair learning with graph regularization for mild cognitive impairment prediction WEI Caifeng,SUN Yongcong,ZENG Xianhua2 (1.College of Computer Science and Technology,Chongqing University of Posts and Telecommunication,Chongqing 400065, China;2.Chongqing Key Laboratory of Computation Intelligence,Chongqing University of Posts and Telecommunications, Chongqing 400065,China) Abstract:Aiming at dictionary pair learning(DPL)methods only consider the reconstruction error of a sub-dictionary from the same class and the sparseness of coefficients from different classes,and do not consider the geometric neigh- borhood topological relationships between images.To improve the classification ability of DPL algorithms,we propose a DPL with graph regularization(GDPL)algorithm based on geometric neighborhood topological relationships.This al- gorithm is based on the idea that keeping the neighborhood relationship makes the distance between the neighborhood projection coefficients of the same kind small,while the distance between projection coefficients of different kinds is large.Experiments on mild cognitive impairment prediction using the ADNII dataset show that the coding coefficient learned from the GDPL algorithm is 2%~6%higher than that which uses the combined biomarker as feature prediction. according to accuracy (ACC)and area under curve (AUC)metrics.Moreover,the experimental result obtained using GDPL is also better than that obtained using DPL algorithm. Keywords:graph regularization;dictionary pair learning;geometric neighborhood relationship;image classification; mild cognitive impairment prediction 目前，多媒体技术飞速发展，图像数量呈指数 级增长，图像分类技术也得到了飞速发展。图像 分类方法主要包括分类器的设计和特征提取两个 收稿日期：2017-09-16.网络出版日期：2018-04-18. 基金项目：国家自然科学基金项目(61672120)：重庆市科委基 部分，目前对图像分类的研究主要集中在分类器 础学科和前沿技术研究一般项目(cstc2015 SjcyjA40036, 性能的改进和改善特征提取方法两方面。稀疏编 cstc2014jcyjA40049). 通信作者：曾宪华.E-mail:zengxh@cqupt.cdu.cn. 码（字典学习）0是图像分类的有效技术之一，用DOI: 10.11992/tis.201709033 网络出版地址: http://kns.cnki.net/kcms/detail/23.1538.TP.20180418.0946.006.html 图正则化字典对学习的轻度认知功能障碍预测 魏彩锋1,2，孙永聪1,2，曾宪华1,2 （1. 重庆邮电大学 计算机科学与技术学院，重庆 400065; 2. 重庆邮电大学 计算智能重庆市重点实验室，重庆 400065） 摘 要：针对字典对学习 (DPL) 方法只考虑了同类子字典的重构误差和不同类表示系数的稀疏性，没有考虑图 像间的几何近邻拓扑关系的问题。通过近邻保持使得在同类近邻投影系数之间的距离较小，而不同类投影系 数之间的距离大，能够有效提高字典对学习算法的分类性能，基于此提出了基于几何近邻拓扑关系的图正则化 的字典对学习 (GDPL) 算法。在 ADNI1 数据集上对轻度认知功能障碍预测的实验表明，使用 GDPL 算法学习 的编码系数作为特征预测的准确率 (ACC) 和 ROC 曲线下的面积 (AUC) 比使用结合生物标志作为特征预测的 准确率提高了 2%~6%，使用 GDPL 算法比 DPL 算法的实验结果也有提高。 关键词：图正则化；字典对学习；几何近邻关系；图像分类；轻度认知功能障碍预测 中图分类号：TP391; R749 文献标志码：A 文章编号：1673−4785(2019)02−0369−09 中文引用格式：魏彩锋, 孙永聪, 曾宪华. 图正则化字典对学习的轻度认知功能障碍预测[J]. 智能系统学报, 2019, 14(2): 369–377. 英文引用格式：WEI Caifeng, SUN Yongcong, ZENG Xianhua. Dictionary pair learning with graph regularization for mild cognit￾ive impairment prediction[J]. CAAI transactions on intelligent systems, 2019, 14(2): 369–377. Dictionary pair learning with graph regularization for mild cognitive impairment prediction WEI Caifeng1,2 ，SUN Yongcong1,2 ，ZENG Xianhua1,2 (1. College of Computer Science and Technology, Chongqing University of Posts and Telecommunication, Chongqing 400065, China; 2. Chongqing Key Laboratory of Computation Intelligence, Chongqing University of Posts and Telecommunications, Chongqing 400065, China) Abstract: Aiming at dictionary pair learning (DPL) methods only consider the reconstruction error of a sub-dictionary from the same class and the sparseness of coefficients from different classes, and do not consider the geometric neigh￾borhood topological relationships between images. To improve the classification ability of DPL algorithms, we propose a DPL with graph regularization (GDPL) algorithm based on geometric neighborhood topological relationships. This al￾gorithm is based on the idea that keeping the neighborhood relationship makes the distance between the neighborhood projection coefficients of the same kind small, while the distance between projection coefficients of different kinds is large. Experiments on mild cognitive impairment prediction using the ADNI1 dataset show that the coding coefficient learned from the GDPL algorithm is 2%~6% higher than that which uses the combined biomarker as feature prediction, according to accuracy (ACC) and area under curve (AUC) metrics. Moreover, the experimental result obtained using GDPL is also better than that obtained using DPL algorithm. Keywords: graph regularization; dictionary pair learning; geometric neighborhood relationship; image classification; mild cognitive impairment prediction 目前，多媒体技术飞速发展，图像数量呈指数 级增长，图像分类技术也得到了飞速发展。图像 分类方法主要包括分类器的设计和特征提取两个 部分，目前对图像分类的研究主要集中在分类器 性能的改进和改善特征提取方法两方面。稀疏编 码 (字典学习) [1-10]是图像分类的有效技术之一，用 收稿日期：2017−09−16. 网络出版日期：2018−04−18. 基金项目：国家自然科学基金项目 (61672120)；重庆市科委基 础学科和前沿技术研究一般项目 (cstc2015jcyjA40036， cstc2014jcyjA40049). 通信作者：曾宪华. E-mail：zengxh@cqupt.edu.cn. 第 14 卷第 2 期 智 能 系 统 学 报 Vol.14 No.2 2019 年 3 月 CAAI Transactions on Intelligent Systems Mar. 2019