面向web视频的数据挖掘及检索的研究和实现

发布时间：2018-04-25 14:33

本文选题：面向web视频检索 + 镜头检测　；参考：《电子科技大学》2012年硕士论文

【摘要】：随着Internet和多媒体技术的发展，使得web上的信息种类繁多，其中尤以视频形式涵盖的信息更丰富生动，它们表现了社会和生活的各个方面。如何让人们能够从这浩如烟海的视频信息中找到自己需要的视频，已经成为目前急需解决的重要问题。传统的对视频进行检索的方法是基于文本的方式，即人工为视频进行文本注释，然后利用传统的信息检索技术对文本进行检索。但由于视频数据具有相当丰富的信息，难以用精确的语言来描述它的特征，因此采用传统的检索方式就存在着弊端。而基于内容的视频检索技术能很好地解决这一弊端。基于内容的视频检索技术，是通过对视频进行镜头检测，然后提取镜头关键帧，对特征进行提取后，利用视频特征进行检索的技术。然而，web上视频数据量大，如何在基于内容的视频检索技术的基础上提供更快捷、有效地视频检索，还必须对视频进行有效地组织和索引，因此对视频信息进行数据挖掘的重要性日益突出。本文以研究基于内容的视频检索为主，对关键技术特别视频镜头检测、关键帧提取、特征提取等技术进行了深入研究，并且研究了数据挖掘的聚类算法，以实现面向web视频数据的数据挖掘。本文以基于关键帧的方法进行视频检索，为了进一步提高检索的效果，在检索前，对关键帧进行预分类。项目组的基于web的搜索引擎系统中对图像进行预分类时采用了人工的方式，将图像分为20类来建立基本图像库。本文提出对关键帧分类的优化算法是不用人工参与，而是由计算机自动分类，即根据底层特征和对应的关键字，采用K均值聚类的方法对关键帧进行分类。经对计算机自动分类的图像库所进行的检索试验表明，与基于web图像搜索引擎系统的人工分类方法相比具有同样高的查准率，这说明了该方法的有效性，克服了人工分类带来的不足。最后，本文设计了一个面向web视频的检索原型演示系统的总体架构，，讨论了系统的模块和功能，并对本文提出的方法进行了实验比较和性能分析。
[Abstract]:With the development of Internet and multimedia technology, there are many kinds of information on web, especially the rich and vivid information in the form of video, which represents all aspects of society and life. How to enable people to find their own video from the vast amount of video information has become an important problem that needs to be solved. The traditional method of video retrieval is based on text, that is, manual text annotation for video, and then using traditional information retrieval technology to retrieve text. However, because video data is rich in information, it is difficult to describe its features in precise language, so there are drawbacks in traditional retrieval methods. Content-based video retrieval technology can solve this problem well. Content-based video retrieval is based on the shot detection of video, then the key frame of shot is extracted, the feature is extracted, and then the video feature is retrieved. However, because of the large amount of video data on the web, how to provide faster and more effective video retrieval on the basis of content-based video retrieval technology must be organized and indexed effectively. Therefore, the importance of video data mining is becoming more and more important. This paper focuses on the research of content-based video retrieval. The key technologies, such as special video shot detection, key frame extraction, feature extraction and so on, are studied deeply, and the clustering algorithm of data mining is also studied. In order to realize the data mining for web video data. In this paper, video retrieval is based on key-frame method. In order to further improve the retrieval effect, the key-frame is pre-classified before retrieval. In the web based search engine system of the project team, the image is classified into 20 categories to establish the basic image database. In this paper, an optimization algorithm for key frame classification is proposed. Instead of manual participation, the algorithm is automatically classified by computer, that is, according to the underlying features and corresponding keywords, the K-means clustering method is used to classify key frames. The retrieval experiment on the image database of computer automatic classification shows that it has the same precision rate as the artificial classification method based on web image search engine system, which shows the effectiveness of this method. It overcomes the disadvantage of manual classification. Finally, this paper designs a prototype demonstration system for web video retrieval, discusses the modules and functions of the system, and makes experimental comparison and performance analysis of the methods proposed in this paper.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.41;TP311.13

【参考文献】