项目名称: 网络环境下基于视觉显著性的图像检索
项目编号: No.61472227
项目类型: 面上项目
立项/批准年度: 2015
项目学科: 自动化技术、计算机技术
项目作者: 华臻
作者单位: 山东工商学院
项目金额: 84万元
中文摘要: 随着互联网技术和多媒体技术的高速发展,如何对网络中大量的图像数据进行有效的管理并从中高效地检索到需要的图像已成为急需解决的问题。面向网络的图像检索不同于面向传统数据库的图像检索,它的数据源是异构的,相关处理算法大多是建立在非线性的、高维数据的基础之上的,存在检索时间长、效率低的问题。 本项目研究网络环境下基于视觉显著性的图像检索技术,构造基于双密度Contourlet变换的理论模型,在Contourlet变换域中采用形变模型进行轮廓提取。改进视觉关注度模型,将颜色特征图、亮度特征图和边缘特征图线性组合成一幅感兴趣区域图。在双密度Contourlet变换域下来研究图像统计特性及显著区域纹理特征,采用一种具有仿射不变性特征的描述子来表达显著区域特征。图像之间有针对性地选择同类别视觉显著性区域进行区域相似性比较,可以更有效地表达图像的主要语义内容,提高图像检索的效率和准确性。
中文关键词: 数据挖掘;资源管理;人工智能
英文摘要: With the rapid development of the Internet and multimedia technologies, there are large amount of image data in the network, and the effective management and efficiently retrieve the required image has become an urgent problem. Network-oriented image retrieval is different from traditional database-oriented image retrieval, the data source is heterogeneous, related processing algorithms are mostly built on non-linear and high-dimensional data, with long search time and low efficiency. The project study the image retrieval based on visual saliency for Network environment. We construct a theoretical model based on double density Contourlet transform, and extract contour using deformable model in Contourlet transform domain. Visual attention model is improved, we map a linear combination of color feature maps brightness feature maps and edge features into an interest region figure. In double density Contourlet transform domain, the image statistical characteristics and significant regional texture features are studied, using an affine invariant feature descriptors to express significant regional characteristics. Selecting the same class the visual salience regional and comparing similarity regional between images, it can more effectively express the semantic content of the image, and improve the efficiency and accuracy of the image search.
英文关键词: Data Mining;Resource Management;Artificial Intelligence