摘要
为解决图像处理中的高维特征在模式分类中带来的问题,提出一种基于半监督学习理论的数据降维方法,称为局部敏感的半监督鉴别分析算法。为能够发现局部的流形结构,算法寻找一个能够最小化类内距离的同时最大化类间距离的投影,并且在最优化过程中充分利用无标签数据,控制局部邻域的散度。在人脸识别数据库和行为数据库中的测试结果表明了该算法是有效的。
To solve high-dimensional feature in the pattern classification problems,a data dimensionality reduction method using semi-supervised learning is presented,called locality sensitive semi-supervised discriminate analysis（LSSDA）.By discovering the local manifold structure for discriminant analysis,LSSDA finds a projection which minimizes the with-class distance while maximize the between-class distance.The unlabeled data is used as regulatory factors during the optimize process.Extensive experimental results on face recognition database and action recognition database demonstrate the proposed algorithm has an encouraging recognition performation.
出处
《计算机工程与设计》
CSCD
北大核心
2010年第10期2283-2286,共4页
Computer Engineering and Design
基金
国家自然科学基金项目(70781043)
