CASIA OpenIR  > 毕业生  > 硕士学位论文
基于内容的视频结构分析
Alternative TitleContent-Based Video Structure Analysis
曾祥林
Subtype工学硕士
Thesis Advisor胡卫明
2009-05-12
Degree Grantor中国科学院研究生院
Place of Conferral中国科学院自动化研究所
Degree Discipline模式识别与智能系统
Keyword视频结构分析 镜头切割 关键帧提取 场景切割 主导集聚类算法 Video Structure Analysis Shot Segmentation Key-frame Extraction Scene Segmentation Dominant-set Clustering
Abstract基于内容的视频结构分析可以大大提高视频的结构化程度,方便用户对视频内容的获取和浏览,也是其他视频相关技术的基础。本文使用机器学习的方法对镜头切割,镜头关键帧提取和场景切割的内容做了相关研究,主要工作体现如下: ① 提出了基于主导集聚类算法的镜头的关键帧提取方法。该方法能自动决定关键帧的个数,能够很好的适应镜头内容的复杂程度;该方法能够渐进的产生层次清晰的关键帧,容易控制产生关键帧的过程。 ② 在基于镜头聚类的场景切割方法中,提出了使用主导集聚类算法进行镜头聚类来实现场景切割的方法。由于使用主导集聚类算法,镜头聚类时事先不需要设定镜头组的个数。场景切割的实验结果表明使用主导集聚类算法进行镜头聚类要优于使用多类Normalized Cut聚类算法。 ③ 在基于边界检测的场景切割方法中,提出了基于Normalized Cut和Min-Max Cut准则构造镜头的连贯性信号的方法。Normalized Cut和Min-Max Cut的准则既考虑了一个场景内的镜头内容上的相似性,又考虑了不同场景的镜头内容的不相似性,所以构造的连贯性信号更好的体现了镜头内容上的连贯性。 ④ 在镜头自动切割,镜头关键帧提取和视频场景的切割的基础上,实现了简易的视频导航系统,方便了用户对视频内容的获取和浏览。
Other AbstractContent-based video structure analysis can improve the degree of video structure greatly, provide convenience to users on browsing the content, and is also the base of other related video techniques. Based on machine learning methods we do some research on shot segmentation, shot-based key-frame extraction and scene segmentation in the thesis. The main contributions of this thesis are summarized as: ①We propose a key-frame extraction approach using dominant-set clustering. Clustering is a popular approach for key-frame extraction. But the number of clusters should be set in advance for general clustering algorithms. Our proposed method dynamically decides the number of key frames depending on the complexity of video shots and produces key frames in a progressive manner. ②We propose a shot clustering based scene segmentation approach using dominant-set clustering. Due to dominant-set clustering algorithm, the number of shot clusters needn't set in advance. Experiments on different video materials show that the dominant-set clustering algorithm achieve better performance than multi-way normalized cut algorithm in scene segmentation. ③We propose a novel method to construct shot coherence signal using Normalized Cut and Min-Max Cut criteria. Then the continuous signal is used to detect scene boundary. Because the Normalized Cut and Min-Max Cut criteria simultaneously emphasizes on the homogeneity between the shots in the same scene and inhomogeneity between the shots in different scenes, so the continuous signals based on the criteria reflect the shot content coherence well. ④We develop a system for video navigation based on shot segmentation, key-frame extraction and scene segmentation. This system provides great convenience to users on browsing the content of videos at a glare.
shelfnumXWLW1411
Other Identifier200628014628063
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/7471
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
曾祥林. 基于内容的视频结构分析[D]. 中国科学院自动化研究所. 中国科学院研究生院,2009.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20062801462806(6746KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[曾祥林]'s Articles
Baidu academic
Similar articles in Baidu academic
[曾祥林]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[曾祥林]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.