基于変分PDE的显著特征提取及其在图像检索中的研究

项目名称： 基于変分PDE的显著特征提取及其在图像检索中的研究

项目编号： No.61202349

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 李梦

作者单位： 重庆文理学院

项目金额： 23万元

中文摘要： 随着数字图像飞速增长，如何从海量数据中检索有效图像信息是研究的重要课题。基于内容的图像检索(CBIR)技术就是解决这一问题的关键技术之一。然而图像低层特征与高层语义间的巨大鸿沟导致了图像检索的困难。本项目围绕CBIR语义鸿沟问题研究基于变分模型的图像显著特征提取。该工作是在我们前期工作（我们的前期工作曾被infrared physics & technology等期刊匿名评审评价具有原创性）基础上的一个新的研究课题。通过对各项异性扩散模型的分析和复值化图像特征分析，本项目拟设计基于复数域的变分模型，通过能量泛函的极小来获取具有高层语义、能够引起人们视觉注意的图像显著特征，该方法为检测图像的显著特征提供了一个新的思路。事实上相似性度量是图像精确检索的另一重要话题，本项目拟建立基于分数阶熵的相似性度量，它是分数阶偏微分方程理论在实际问题中一个新的尝

中文关键词： 显著性检测；特征提取；视觉注意；变分模型；偏微分方程

英文摘要： With the substantial increasing number of digital images, it has become an important issue to retrieve required information from massive image data efficiently and rapidly. Content based image retrieval (CBIR) is just one of key technologies for such a problem. However, semantic gap, difference between visual features and semantic annotations, is a problem of CBIR systems. This project makes a research on saliency detection based on variational model while overcoming semantic wide gap on CBIR. This makes our method and the existed models of saliency detection are quite different. Our work takes benifits of our previous works , which are highly commended by many reviewers.(For example: review of Infrared Physics & Technology reported, it is an interesting idea and important contribution concerning infrared image segmentation. The proposed method is somehow original.) We analyze an anisotropic diffusion model and then extend some variational models by considering image features in the complex field. Various features for visual attention can be detected by minimizing the energy functional. During the past decades many researchers have devoted to the development of variational models and proposed, many good algorithms to solve important topics in image analysis and computer vision. However, to our knowledge, there e

英文关键词： Sailency detection；Features extraction；Visual attention；Variational model；Partial differential equation