广告视频探测技术研究
发布时间：2018-11-17 07:40
【摘要】：随着市场经济的深入发展,视频广告对于企业越发重要,并逐渐成为企业日常工作的一部分。广告公司和企业为了确保所做的视频广告取得应有的效益,必须派人来监测电视台对于合同的执行情况。同时文化管理部门会依据政府的行政命令要求电视台播出一些公益公告,政府通告、通知等。为了确保这些强制插播的视频在规定的时间内完全播放,文化监管部门也必须指定专人进行检测。目前视频广告的监测都是通过人工进行的,浪费了大量的人力物力资源。 视频数字化浪潮的发展使得电视台和信息服务商对于视频分类技术的依赖越来越重。广告视频探测作为视频分类的一个分支,其自动探测的实现为其他类型的视频探测提供借鉴。 本文探讨了广告视频的特点和实现广告视频自动探测的思想方法,把广告视频自动探测分为两个方面:广告板块探测和广告单元探测。针对广告板块探测,提出了基于多特征融合的滑窗边界探测算法;针对广告单元的检测,提出了基于帧匹配和镜头长度序列匹配相结合的广告单元探测算法。 本文的主要工作在于: ● 分析了广告视频的特点及其对应于视频结构、视频对象、音频等底层的特征,提出了广告视频的探测技术框架。 ● 提出了适用于广告板块边界探测的基于多特征融合的滑窗视频边界探测算法。研究了获取广告视频特征向量各分量的提取技术,这些技术包括压缩域的镜头分割技术、字幕区域探测技术、特征向量在某一置信度下的近似匹配技术。分析了表征广告视频类型属性的特征向量的各分量的权重的分配。 ● 提出了适用于广告单元的帧匹配与镜头相似匹配相结合的视频段匹配算法。探讨了基于DC分量的帧匹配算法和基于镜头长度序列的镜头相似匹配的视频段匹配算法。 ● 在上述技术的基础上设计并实现了广告视频探测的系统,验证了上述各章的思想方法。
[Abstract]:With the further development of market economy, video advertising is becoming more and more important for enterprises, and gradually become a part of the daily work of enterprises. Advertising agencies and companies must send people to monitor the implementation of contracts by television stations in order to ensure that video advertising is paid for. At the same time, according to the administrative order of the government, the cultural administration will require the television station to broadcast some public service announcements, government circulars, notices, etc. In order to ensure that these mandatory episodes are fully broadcast within the specified time, the cultural regulator must also appoint a dedicated person to conduct the tests. At present, the monitoring of video advertising is carried out manually, wasting a lot of human and material resources. With the development of video digitization, TV stations and information service providers rely more and more on video classification technology. Advertising video detection as a branch of video classification, its automatic detection provides reference for other types of video detection. This paper discusses the characteristics of advertising video and the method of realizing the automatic detection of advertising video. The automatic detection of advertising video is divided into two aspects: ad block detection and advertising unit detection. A sliding window boundary detection algorithm based on multi-feature fusion is proposed for advertising block detection, and an advertising unit detection algorithm based on frame matching and shot length sequence matching is proposed for advertising unit detection. The main work of this paper is to analyze the characteristics of advertising video and its corresponding features such as video structure, video object, audio and so on, and put forward the detection technology framework of advertising video. A sliding window video boundary detection algorithm based on multi-feature fusion is proposed for advertising block boundary detection. In this paper, the extraction techniques of each component of feature vector in advertising video are studied. These techniques include shot segmentation in compressed domain, subtitle region detection and approximate matching of feature vectors under certain confidence. The weight distribution of each component of the feature vector which characterizes the attribute of the video type of advertisement is analyzed. A video segment matching algorithm combining frame matching and shot similarity matching is proposed. Frame matching algorithm based on DC component and video segment matching algorithm based on shot length sequence are discussed. Based on the above technology, the system of advertisement video detection is designed and implemented, and the ideas and methods of the above chapters are verified.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2004
【分类号】：TN948.1
本文编号：2336993
[Abstract]:With the further development of market economy, video advertising is becoming more and more important for enterprises, and gradually become a part of the daily work of enterprises. Advertising agencies and companies must send people to monitor the implementation of contracts by television stations in order to ensure that video advertising is paid for. At the same time, according to the administrative order of the government, the cultural administration will require the television station to broadcast some public service announcements, government circulars, notices, etc. In order to ensure that these mandatory episodes are fully broadcast within the specified time, the cultural regulator must also appoint a dedicated person to conduct the tests. At present, the monitoring of video advertising is carried out manually, wasting a lot of human and material resources. With the development of video digitization, TV stations and information service providers rely more and more on video classification technology. Advertising video detection as a branch of video classification, its automatic detection provides reference for other types of video detection. This paper discusses the characteristics of advertising video and the method of realizing the automatic detection of advertising video. The automatic detection of advertising video is divided into two aspects: ad block detection and advertising unit detection. A sliding window boundary detection algorithm based on multi-feature fusion is proposed for advertising block detection, and an advertising unit detection algorithm based on frame matching and shot length sequence matching is proposed for advertising unit detection. The main work of this paper is to analyze the characteristics of advertising video and its corresponding features such as video structure, video object, audio and so on, and put forward the detection technology framework of advertising video. A sliding window video boundary detection algorithm based on multi-feature fusion is proposed for advertising block boundary detection. In this paper, the extraction techniques of each component of feature vector in advertising video are studied. These techniques include shot segmentation in compressed domain, subtitle region detection and approximate matching of feature vectors under certain confidence. The weight distribution of each component of the feature vector which characterizes the attribute of the video type of advertisement is analyzed. A video segment matching algorithm combining frame matching and shot similarity matching is proposed. Frame matching algorithm based on DC component and video segment matching algorithm based on shot length sequence are discussed. Based on the above technology, the system of advertisement video detection is designed and implemented, and the ideas and methods of the above chapters are verified.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2004
【分类号】：TN948.1
【相似文献】
相关博士学位论文 前1条
1 赵亚琴;基于内容的视频片段检索技术研究[D];南京理工大学;2007年
相关硕士学位论文 前3条
1 李春亮;广告视频探测技术研究[D];国防科学技术大学;2004年
2 张海勇;基于片花的新闻视频搜索引擎的设计和实现[D];河海大学;2006年
3 杨柳;基于内容的流媒体中的视频检索[D];北京科技大学;2010年
，本文编号：2336993
本文链接：https://www.wllwen.com/wenyilunwen/guanggaoshejilunwen/2336993.html
