权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

Edit and Development of Image Information Recommendation System According to Sensibility Information

基于情感信息的图像信息推荐系统的编辑与开发

基本信息

批准号：
16500122
负责人：
WATANABE Makoto
金额：
$ 2.3万
依托单位：
Chiba University
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (C)
财政年份：
2004
资助国家：
日本
起止时间：
2004 至 2005
项目状态：
已结题

来源：
https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-16500122/
关键词：
Program Information Compressed Expression BGM Shot Cut Serif Volume

项目摘要

In this paper, automatic generation of summary image for program selection to assist audience who watch TV drama is proposed. Nowadays, along with the initiation of digital terrestrial broadcasting and popularization of hard dick DVD recorder, TV viewing audience have more opportunities to record, manage, and use a lot of image data. In the future, they will be in a difficult situation to screen the programs to watch or information to get among the vast amount of program contents. Summary image to assist them would be needed. Manual works such as indexing input or meaning of scene were required in many of the traditional study for image summary. In this study, specific data only (volume, BGM, the length of shot) in drama image is used in order to promote the further automation.As the purpose is to assist quick program selection, instead of connecting main sections sampled to fix the total length, it is designed that the summary image would consist of 5 seconds x 5 sections (total length : approximately 25 seconds). Reasonable 5-second section is sampled based on the specific data. Among the section (conversation) which the chance of the maximum volume is the most in 5 seconds, section (presentation by BGM) before and after the starting and ending point of BGM, and section (intense scene) which the change of the scene is the most in 5 seconds, the high rankings are selected. When the sections sampled are connected, the cutting point of the voice should be handled very carefully. The cutting points are changed to the minimum value of the quantity shift graph to avoid the voice stopped awkwardly. Practical drama scenes are adopted for sampling evaluation.

本文提出了一种自动生成摘要图像进行节目选择的方法，以帮助观看电视剧的观众。如今，随着数字地面广播的兴起和硬碟DVD刻录机的普及，电视观众有更多的机会记录、管理和使用大量的图像数据。今后，他们将很难在海量的节目内容中选择要看的节目和要获取的信息。将需要概要图像来协助他们。在传统的图像摘要研究中，很多都需要手工的工作，如索引输入或场景含义等。在本研究中，仅使用戏剧图像中的特定数据（体积，BGM，镜头长度），以促进进一步的自动化。为了方便快速选择节目，我们设计的摘要图像由5秒x 5段组成（总长约25秒），而不是将采样的主要部分连接起来固定总长度。根据具体数据选取合理的5秒段。在5秒内音量最大机会最大的片段（对话）、BGM开始和结束点前后的片段（BGM演示）、5秒内场景变化最大的片段（激烈场景）中，选取排名较高的部分。当采样的部分连接时，声音的切割点应处理得非常小心。将切割点更改为量移图的最小值，以避免声音笨拙地停止。采用实际戏剧场景进行抽样评价。