冀中, 熊凯琳, 马亚茹, 何宇清
AuthorsHTML:冀中, 熊凯琳, 马亚茹, 何宇清
AuthorsListE:Ji Zhong, Xiong Kailin, Ma Yaru, He Yuqing
AuthorsHTMLE:Ji Zhong, Xiong Kailin, Ma Yaru, He Yuqing
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:如何有效地管理和查询海量视频数据是大数据时代亟待解决的问题．基于查询的多视频摘要技术可提供全面且简洁的查询内容的相关信息, 是解决此问题的重要途径之一．然而, 多视频内容具有多样性, 且包含较多的噪音和冗余, 从这些复杂信息中找出最具代表性的信息极具挑战性．针对这一挑战, 提出一种基于稀疏自编码器, 并将网络查询图像内容作为正则项的多视频摘要模型．该模型不仅满足代表性和简洁性的要求, 还具有依赖查询进行重要性感知的能力．大量的实验验证了本文模型的有效性与先进性.
Abstract_English:How to manage and search massive video data effectively is an urgent problem in the era of big data. Query based multi-video summarization can provide comprehensive and concise information about the content of query videos，which is one of the promising ways to address this problem. However，the content of multiple videos is diverse，noisy and redundant，which makes it very challenging to find the most representative information from these videos. A sparse auto-encoder-based multi-video summarization model is proposed，using web query images as regularization terms. It not only satisfies the criteria of representativeness and conciseness，but also has the capability to perceive the query-dependent importance. Extensive experiments demonstrate its effectiveness and superiority.
Keyword_Chinese:多视频摘要; 稀疏自编码器; 重要性感知; 视频管理
Keywords_English:multi-video summarization; sparse auto-encoder; importance awareness; video management

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=6129