项目名称: 海量可视媒体信息管理的索引技术研究
项目编号: No.61300189
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 齐恒
作者单位: 大连理工大学
项目金额: 27万元
中文摘要: 随着图像、视频之类的可视媒体信息数量与日俱增,海量可视媒体的管理已成为热门研究领域。在海量可视媒体管理的研究中,如何构建合理的可视媒体索引机制属于核心问题,是优化海量可视媒体信息检索等多项处理技术的基础。 本项目将针对可视媒体索引机制展开研究工作。工作中首先对视觉词包索引,该项最具代表性的局部信息索引机制进行研究,分别从特征空间的选择,视觉词表的构建,视觉短语的定义三个方面来对现存问题进行分析,找出解决方案,提出具有自主创新性的基于视觉单词与视觉短语相融合的视觉词典索引机制。其次,针对局部信息索引机制所存在的缺陷,研究如何在模态探索技术基础上构建可视媒体的全局信息索引,并通过对其扩展形成一套全新的基于多层模态探索的复合型索引机制。最后,搭建海量可视媒体检索平台,以及相应的性能评测系统,对项目研究成果进行统一验证。
中文关键词: 可视媒体管理;索引机制;视觉词包;可视媒体计算;体系架构
英文摘要: With the explosive growth of visual media including images and videos, the management of massive visual media becomes a hot topic in the research fields. In the management of massive visual media, it is a key problem how to build effective indexing mechanism for optimizing the massive visual media retrieval and other processing technologies. In this project, we focus on the indexing mechanism of massive visual media. Firstly, we find problems about the existing local information indexing based on bag-of-visual-words (BOVW) from three respects: the selection of feature space, the building of visual vocabulary and the definition of visual phrase. To address these existing problems, we propose a new visual dictionary indexing consisting of visual words and visual phrase. Secondly, to overcome the defects in the local information indexing, we focus on the problem how to build the global information indexing based on the mode-seeking technology. According to research about mode-seeking, we propose a new compound indexing mechanism based on multilayer mode-seeking. Finally, we build the experiment platform of visual media information retrieval and the performance analyzing system to verify our research results.
英文关键词: Visual Media Management;Indexing Mechanism;Bag-of-visual-word;Visual Media Computing;Architecture