冀中, 郭威辰
AuthorsHTML:冀中, 郭威辰
AuthorsListE:Ji Zhong, Guo Weichen
AuthorsHTMLE:Ji Zhong, Guo Weichen
Unit:天津大学电气自动化与信息工程学院，天津 300072
Unit_EngLish:School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
Abstract_Chinese:动作识别领域需要识别的类别越来越多, 这使得标注足够多的训练数据越来越难．零样本学习是针对传统机器学习收集和标注数据日益困难而提出的一种新思路．针对基于零样本学习的动作识别问题, 提出了一种基于局部保持典型相关分析映射的方法．该方法使用流形约束的典型相关分析将视觉特征和辅助特征映射到一个公共特征空间, 并且在映射过程中保留视觉特征和辅助特征的局部信息, 还考虑了域转换所带来的不利影响, 同时采用自训练和hubness修正等方法增强所提方法的鲁棒性．通过在主流数据集HMDB51和UCF101上的大量实验, 表明所提方法具有较好的零样本学习性能．
Abstract_English:The number of categories for action recognition is growing rapidly and it has become increasingly hard to label sufficient training data for learning classification models of all categories. Zero shot learning(ZSL)is an attractive approach aiming at handling the difficulty in collecting ever more data and labeling them exhaustively. This paper proposes a ZSL-based action recognition method with the idea of local preserving canonical correlation analysis(LPCCA). Specifically，a mapping from visual and side information to a common CCA feature space is constructed，using a manifold-regularized term. The impact of domain shift is also taken into consideration. Approaches of self-training and hubness correction are applied to improve the robustness of the proposed method. The proposed method is evaluated extensively on popular human action datasets of HMDB51 and UCF101. The results demonstrate that the proposed method achieves a better performance against the state-of-the-art with a simple and efficient pipeline.
Keyword_Chinese:零样本学习; 动作识别; 典型相关分析; 局部保持
Keywords_English:zero shot learning(ZSL); action recognition; canonical correlation analysis(CCA); local preserving

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=5914