融合视觉内容分析的网络视频缩略图推荐  

Web Video Thumbnail Recommendation by Visual Content Analysis

作　　者：张维刚[1] 王祯骏 黄庆明[2] 高文[3] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001 [2]中国科学院大学计算机与控制学院,北京100049 [3]北京大学信息科学技术学院,北京100871

出　　处：《智能计算机与应用》2014年第3期5-10,共6页Intelligent Computer and Applications

基　　金：国家重点基础研究发展计划(973)(2012CB316400);国家自然科学基金(61025011;61332016;61202322)

摘　　要：视频在网站上呈现时,最先映入眼帘的都是一幅静态的封面图像——视频缩略图。主题明确且图像易见的视频缩略图,能让用户在预览时更容易、更快捷地抓住视频主题信息,提高该视频的吸引力,获得更多的点击量和提升用户的浏览体验及观看兴趣。本文针对网络视频缩略图的推荐,在视觉内容分析基础上,通过融合图像易获取性评价和视频内容代表性评价来获得缩略图推荐结果。在网络视频数据集上实施的主客观评价实验表明,本文方法所推荐的视频缩略图与原有人工标注的缩略图相比,有着较好的一致性,能够满足实际应用的需求。When a web video is shown on the website, its static cover image, which is also called video thumbnail, will firstly come into sight. A good thumbnail, which is easy to be accessed and much relevant to the video theme, can give the browsers as much content iniormation as possible at a glance and makes the video more attractive. It will bring more website traffic and the user＇ s video browsing experience will also be improved. To attempt to solve the video thumbnail recommen- dation problem, the paper proposes a new approach based on visual content analysis of video keyframes. This method in- cludes three operations： image accessibility evaluating, video content representativeness evaluating and fusing. The subjec- tive and objective experiments on the web video dataset show good consistency between the recommended thumbnails and the original manual ones, which shows that the presented framework can meet the actual needs.

关 键 词：视频缩略图 图像易获取性 视频内容代表性 支持向量回归 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]