CASIA OpenIR  > 毕业生  > 硕士学位论文
Alternative TitleAn Interactive Video Retrieval Framework Using Semantic Matching and Information Fusion
Thesis Advisor胡卫明
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计算机应用技术
Keyword视频特征 检索机制 相关反馈 信息融合 Feature Extraction Retrieval Mechanism Relevance Feedback Information Fusion
Abstract随着多媒体技术和互联网技术的飞速发展,以视频为代表的多媒体数据正在以惊人的速度增长。面对如此丰富、无序、海量的多媒体数据,如何实现所需资源的有效组织、高效检索和快速获取已成为人类社会面临的巨大挑战,同时也使多媒体检索技术迅速成为当今最热门的研究领域之一。 基于内容的视频检索(CBVR)是多媒体检索领域的重要分支,它摆脱了人工文本标注的传统方式,直接对视频所蕴涵的物理和语义内容进行分析与理解以达到快速准确的检索效果。本文通过对高层语义特征、视频检索机制、相关反馈算法和多源信息融合等四个方面的研究,提出一套新型的基于基本语义的CBVR系统——“基于主题匹配与信息融合的交互式视频检索框架”,其主要贡献包括: (1) 提出基于模型匹配策略和主题匹配策略的视频特征提取方法。定义新型中层特征——模型匹配相关图用以精确描述视频序列的时空信息;定义新型高层特征——主题直方图用以实现语义关键词的自动标注和对视频基本语义内容的表征。 (2) 提出基于非监督学习的视频检索机制。使用Dominant Set聚类算法,建立基于非监督学习的索引结构和基于聚类索引的查询机制,显著降低了检索系统的在线计算复杂度同时实现了对数据库更为有效的管理。 (3) 提出新型的视频检索相关反馈技术——基于语义的分支反馈算法。该算法采用分支反馈结构和分支更新策略,通过在线补偿监督信息来校正主题直方图所标记的不恰当的语义关键词,进而显著提升系统的检索性能。 (4) 建立基于颜色信息和运动信息融合的视频检索模式。使用光流分析方法描述视频帧间的局部运动信息,并将其作为视频的底层图像特征嵌入主题匹配策略生成基于运动信息的主题直方图; 使用Dempster-Shafer证据理论将基于颜色信息的主题直方图和基于运动信息的主题直方图进行融合,从而实现基于多源信息融合的视频检索。 (5) 根据上述四部分研究,设计并实现了“基于主题匹配与信息融合的交互式视频检索原型系统”——“SMIF VideoSearch 系统”。
Other AbstractWith the rapid growth of the technology of multimedia and network, the amounts of multimedia data (such as image and video) are increasing greatly. How to effectively organize, store and retrieve the huge data is challenging us and making the multimedia retrieval become one of the hottest research fields. Content-based video retrieval (CBVR) as a branch of the multimedia retrieval has attracted increasing interests, for it directly relies on video content in contrast to traditional retrieval techniques using manual annotations. In this thesis, we focus on the following four parts: video feature extraction, retrieval mechanism, relevance feedback and information fusion to propose an original CBVR system named “An Interactive Video Retrieval Framework using Semantic Matching and Information Fusion”. The main contributions of this thesis are five-fold. (1) We propose a novel feature extraction method based on the model matching and semantic matching strategy. A new mid-level sequence feature named “Model- Matching Correlogram” is extracted to accurately describe video’s spatio-temporal information, while a new high-level semantic feature called “Semantic-Matching Histogram” is defined in order to uncover videos’ basic semantic content. (2) We establish an unsupervised learning-based retrieval mechanism, which consists of indexing process and querying process, using the Dominant Set clustering for the sake of low on-line complexity and high retrieval efficiency. (3) We develop a new relevance feedback algorithm called Semantic-Based Relevance Feedback working together with SMHs to correct the inaccurate semantic keywords labeled by SMHs and improves the retrieval performances remarkably. (4) We set up a video retrieval method based on the fusion of color and motion information. Optical flow analysis is used to reflect the local motion between adjacent frames and embedded into our semantic-matching strategy, while Dempster-Shafer theory is imported to fusion the SMHs on Color and SMHs on Motion. (5) We design and implement the interactive video retrieval prototype system using semantic matching and information fusion —— “SMIF VideoSearch system”.
Other Identifier200528014628038
Document Type学位论文
Recommended Citation
GB/T 7714
李华北. 基于主题匹配与信息融合的交互式视频检索框架[D]. 中国科学院自动化研究所. 中国科学院研究生院,2008.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_20052801462803(4266KB) 暂不开放CC BY-NC-SA
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[李华北]'s Articles
Baidu academic
Similar articles in Baidu academic
[李华北]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[李华北]'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.