项目名称: 压缩感知集成分类器设计研究
项目编号: No.61272052
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 张宝昌
作者单位: 北京航空航天大学
项目金额: 79万元
中文摘要: 人脸识别是模式识别领域的经典研究问题,具有重要的理论意义和广阔的应用前景。申请人成功地完成了国家自然科学基金青年基金项的主要研究内容,发表了6篇国际期刊论文以及多个著名的国际会议论文,其中包括了IEEE Trans. Image Processing,IEEE Trans. CSVT 及CVPR。在特征表示及分类器设计方面进行了创新。 高性能分类器设计是人脸及其他目标识别的核心关键问题之一,受到模式识别、机器学习领域的广泛重视。本课题主要研究新型集成分类器的构建,拟引入压缩感知理论、VC维最小化、相关性最小化等概念进行分类器设计。还将探讨基于迁移学习的分类器设计方法,使得分类器能够适应测试集合样本分布的变化,为实际应用提供理论支撑。本课题的成果将在人脸检测、眼睛定位、人脸识别等上进行应用,并经过大量的实验验证,从而进行系统集成来构建高效的人脸识别系统。
中文关键词: 集成分类器;在线学习;目标跟踪;;
英文摘要: Face Recognition is a research challenge in Pattern recognition and has important theoretical significance and practical applications. In the previous project sponsored by NSFC, we successfully complete the original plan, and publish 6 distinguished international journals with 2 top ones, as well as several excellent conference papers (such as CVPR). We did innovative work on feature representation and classifier design. This proposal is mainly focusing on the problem of classifier design. We plan to introduce: 1) compressed sensing theory, 2) VC-dimension minimization, and 3) relevance minimization towards optimal classifier construction. We also plan to employ the transfer learning theory which can enable the training set derived classifier adaptive to the distribution variation of the test set, as well as the parameterized classifier fusion approach which can further improve the classifier performance. The output of this proposal will be applied to face detection, eye localization, and face recognition. Through extensive experimental justification and system integration, a highly effective and robust face recognition system is expected to be implemented.
英文关键词: ensemble;online learning;tracking;;