基于多证据融合的视频排序方法

发布时间：2018-05-20 18:10

  本文选题：多证据融合 + 视频排序　； 参考：《电子学报》2010年01期

【摘要】：在视频检索中,通过对用户行为特性的分析发现,用户通常只关注排在最前面的返回结果,而很少有耐心将所有的返回结果浏览一遍.因此,对于一个搜索引擎来说,能否将最相关的结果排在最前面是至关重要的.为了实现这一目标,本文提出了一种基于多证据融合的视频排序方法.该方法利用Dempster-Shafer证据推理理论来协同地融合多方证据,进而推断出最相关的视频镜头.如果多方证据一致,则证明某个视频镜头是相关的,此镜头被认为是最相关的镜头,并被排在返回列表的最前列.相反,如果多方证据产生冲突,那么此镜头就将被排在后面.实验结果表明,利用建议的多证据融合排序算法,搜索引擎的搜索质量,特别是排在前列的搜索结果的准确性,有了明显的改善.
[Abstract]:In video retrieval, by analyzing the behavior of users, it is found that users usually focus on the first result, and rarely have the patience to browse all the returned results once. Therefore, for a search engine, whether the most relevant results at the top is crucial. In order to achieve this goal, a video sorting method based on multi-evidence fusion is proposed. The method uses Dempster-Shafer evidence reasoning theory to fuse multi-party evidence and then infer the most relevant video shot. If the evidence is consistent, it proves that a video shot is relevant, which is considered the most relevant shot and is at the top of the return list. On the contrary, if multiple sources of evidence conflict, then this scene will be placed in the back. The experimental results show that the search quality of search engines, especially the accuracy of the first search results, has been improved obviously by using the proposed multi-evidence fusion sorting algorithm.
【作者单位】： 北京交通大学信息科学研究所;
【基金】：国家自然科学基金(No.60602030,60776794) 国家863高技术研究发展计划(No.2007AA01Z175) 国家重点基础研究发展计划(No.2006CB303104) 教育部长江学者和创新团队发展计划(No.IRT707) 模式识别国家重点实验室开放基金
【分类号】：TP391.41
，

本文编号：1915700