An Automatic Video Reinforcing System Based on Popularity Rating of Scenes and Level of Detail Controlling
An Automatic Video Reinforcing System Based on Popularity Rating of Scenes and Level of Detail Controlling
复制标题
DOI：
10.1109/ism.2015.31
复制
发表时间：
2015-12
期刊：
影响因子：
--
通讯作者：
Yuanyuan Wang;Yukiko Kawai;K. Sumiya;Y. Ishikawa
中科院分区：
文献类型：
--
作者：
Yuanyuan Wang;Yukiko Kawai;K. Sumiya;Y. Ishikawa
随着诸如Netfix的视频点播（VOD）服务的发展，用户能够随时随地观看多种视频。在观看视频的同时，近来，用户经常使用移动的PC通过Web搜索关于视频的相关信息。然而，当用户搜索视频时，视频一直在播放，用户不能满意地理解和欣赏视频，因此需要检测视频的各种问题以补充关于每个场景的相关信息以用于自动搜索。然而，只有一个视频包括每个场景的各种主题，而且，观众的知识水平不同。因此，我们开发了一种新型的自动视频增强系统，称为TV-Binder，它通过添加其他相关内容（即，YouTube视频，图像或地图），并根据每个场景的主题删除不必要的原始场景。结果，观看者可以满意地和愉快地观看修改的视频内容，而无需搜索任何内容。首先，我们的系统提取的主题和检测他们的场景的视频流通过使用隐藏字幕。然后，系统搜索其他必要的内容，并确定不需要的原始场景的基础上，每个原始场景的流行度和细节水平（LOD）的时间压力下控制。通过这种方式，TV-Binder可以自动生成视频内容，并通过两个轴将其分为四个象限，一个是摘要和详细的视频，另一个是针对具有特定主题知识的专家和普通观众的视频。在本文中，我们讨论了我们的自动视频增强系统和评估其有效性。
With the advance of video-on-demand (VOD) services such as Netfix, users are able to watch many kinds of videos anytime and anywhere. While watching a video, recently, users often search related information about it through the Web by using mobile PC. However, users cannot satisfactorily understand and enjoy it because the video keeps playing when they search about it. It is necessary to detect various questions of the video to supplement their related information about each scene for automatic search. However, only one video includes various topics of each scene, furthermore, viewers have different levels of knowledge. Therefore, we have developed a novel automatic video reinforcing system, called TV-Binder, it generates new video contents from one video stream related to viewers' interests and knowledge by adding other related contents (i.e., YouTube videos, images or maps) and by removing unnecessary original scenes, based on topics of each scene. As a result, viewers can satisfy and joyfully watch modified video contents without searching anything. At first, our system extract topics and detect their scenes of a video stream by using closed captions. The system then searches other necessary contents and determines unwanted original scenes based on popularity rating of each original scene and level of detail (LOD) controlling under time pressure. Through this, TV-Binder can automatically generate video contents are classified into four quadrants by two axes, one is digest and detailed videos, the other one is videos for experts with knowledge about particular topics and ordinary viewers without special knowledge. In this paper, we discuss our automatic video reinforcing system and an evaluation of its effectiveness.