基于定性空间推理的空间模式发现及机器视觉应用

项目名称： 基于定性空间推理的空间模式发现及机器视觉应用

项目编号： No.61472161

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 其他

项目作者： 王生生

作者单位： 吉林大学

项目金额： 82万元

中文摘要： 图像中可以提取兴趣点(区域)、边界线等图像基元，这些图像基元之间存在较为复杂的空间模式，其中可能隐含了重要的语义信息。然而，由于其空间关系种类繁多、错综复杂，且存在着相互约束，而此前的研究采用较简单的空间关系模型，未能有效发现图像中的深层、细节空间模式。此外，复杂形状和复杂内部结构分析、复杂交互行为与复杂群体行为学习等研究也涉及复杂空间模式发现问题，并且未能很好地解决。通过理论研究与实验发现，将定性空间推理与机器视觉研究有机结合，有望解决该问题。为此，本项目拟在现有定性空间推理研究基础之上，进一步开展针对机器视觉的定性空间推理扩展研究工作，具体包括多元、异构、多粒度、精细空间关系模型，复杂内部结构模型，复杂交互轨迹定性建模，基于空间推理的复杂空间模式发现等。进而与兴趣点检测、词袋模型、概率图模型等机器视觉与模式识别研究相结合，解决复杂场景下的图像检索分类、复杂行为识别等机器视觉难题。

中文关键词： 定性空间推理；知识工程；知识发现；机器视觉；模式识别

英文摘要： Generic image primitives can be extracted from the original images, such as interest points and boundary segments. The relationships between these primitives show complex spatial relation pattern, and even imply some important semantic information. However, previous researches always employ relatively simple spatial relation model and cannot effectively find deep and refined spatial pattern, due to various kinds, complexity and mutual constraint of the relationships between primitives. Additionally, the researches on complex shape and structure analysis, complex interaction behavior learning, and complex group behavior learning also involve the task of spatial pattern discovery, and it is still a challenge need be solved. However, these problems are expected to be solved by adequately combining the researches of Qualitative Spatial Reasoning (QSR) and Computer Vision, through our theory analysis and example finding. Therefore, this project will develop further study on Qualitative Spatial Reasoning which aims at improving methods in Computer Vision, including multi-aspect, isomerous and multi-granularity spatial relationship model, refined spatial relationship model, complex structure model, qualitative description of complex interactive trajectories and complex spatial pattern discovery based on QSR. Then we combine these spatial models with the issues of computer vision, such as interest point detection, bag-of-word model and probabilistic graphical model, in order to solve the tasks of image retrieval, image classification and complex behavior recognition under the complex scenes.

英文关键词： Qualitative Spatial Reasoning;Knowledge Engineering;Knowledge Discovery;Computer Vision;Pattern Recognition