视觉显著协方差矩阵在视频近似拷贝检测中的应用研究

项目名称： 视觉显著协方差矩阵在视频近似拷贝检测中的应用研究

项目编号： No.61300205

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： 郑立刚

作者单位： 广州大学

项目金额： 22万元

中文摘要： 基于内容的图像视频近似拷贝检测是一项旨在检测具有相似来源且重复出现的多媒体的技术，近年来受到人们的重视。但是检测经过各种变换的图像和视频，是一项具有挑战性的工作，原因在于需要提取能够抵抗多种变换的具有鲁棒性和区分性较强的特征描绘子并设计适合相关特征的索引结构。目前，在这一领域，传统的线性特征提取方法占主导地位，但是这些方法或多或少都有其局限性。在本项目中，我们研究基于视觉显著协方差（后文中的SPD矩阵）的非线视频性特征提取方法，并研究其在拷贝检测中的快速搜索问题。围绕提高基于内容的近似视频拷贝检测的检测准确性和检测效率两个目标，我们研究两大内容：1）融合视觉显著和协方差的视频本征非线性表达；2）SPD矩阵的高效近邻搜索技术研究。围绕两大内容，我们需要解决4个关键问题：1）高效视频显著协方差的计算；2）视觉显著特征的界定和抽取；3）SPD矩阵的高效相似性计算问题；4）黎曼空间中的高效索引技术

中文关键词： 词汇森林；竞争学习；频率敏感；均匀采样；合同变换

英文摘要： Owing to advances in digital multimedia processing technology and broadband internet access, and the increasing popularity of online media sharing, a huge amount of multimedia (especially images and videos) has been flooding websites , the easy availability of multimedia contents has pleased the public but, they have also created problems such as copyright infringements and wasteful usage of storage space and network bandwidth. A promising approach to tackling this problem is the so-called content-based copy detection (CBCD) approach. A key issue to the successful detection of a copied video or image lies in the design of an effective image or video content descriptor and the suitbale fast detecting algorithm. These descriptors are expected to be both robust and discriminative. Presently, these descriptors are linear, which are not successful in in many cases. This proposal is to study the salient covariance (SPD matrix) as a nonlinear feature for content based near-duplicate video detection. what's more, we will further study the fast searching algorithm for SPD. All in all, we will study two main contents. The first is to study the nonlinear representation for video contents using salient covariance. The second is to study the fast searching algorithm for SPD. This proposal will focus on four key problems as f

英文关键词： Vocabulary Forest；Competitve Learning；Frequency Sensitive；Uniform sampling；Congruent Trasnformation