摘要
将视频集看成Grassmann流形上的子空间集合,结合半监督的拉普拉斯特征映射算法,即基于子空间相似性度量和具有标记子空间的类别信息,将视频集非线性地映射到低维欧氏空间,提出Grassmann流形上半监督特征映射算法对视频目标进行识别,该算法分别在步态视频数据库、人手姿势视频数据库和物体姿势视频数据库上进行了目标识别实验,并和典型的基于子空间相似性的分类算法的识别结果进行对比,证明该算法具有较好的性能。
This paper considers the set of videos to subspaces, and uses semi-supervised laplacian eigenmap which is based on metric of the similarity between subspaces and the classes information of labeled subspaces which could nonlinear map the video set to a low dimensional euclidean space. This paper proposed a novel method which is called semi-supervised feature mapping algorithm on Grassmann manifold to recognize video object. Compared with several typical subspaee-based similarity classification algorithms, the results of experiments based on gait video database, ETH-80 gesture video database and hand-gesture video database show that the proposed method can obtain the best performance.
出处
《重庆邮电大学学报（自然科学版）》
CSCD
北大核心
2014年第2期265-270,共6页
Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)
基金
国家自然科学基金(61075019
61100113)
重庆市自然科学基金(CSTC
2010BB2406)~~
