复杂图形数据内容理解与检索技术研究

项目名称： 复杂图形数据内容理解与检索技术研究

项目编号： No.61003113

项目类型： 青年科学基金项目

立项/批准年度： 2011

项目学科： 金属学与金属工艺

项目作者： 苏丰

作者单位： 南京大学

项目金额： 7万元

中文摘要： 对复杂图形数据内容及其中对象的分析、识别与检索是模式识别和计算机视觉领域的一个重要研究方向，也是图形信息自动化和智能化处理应用的关键支撑技术。本课题重点围绕复杂图形上下文中的图形对象的有效特征描述和组织表示形式开展研究，分别探索了多层次、多分辨率、复合统计和结构特征等适应不同干扰和变换的图形特征描述构造方法，在其基础上进一步研究了复杂图形上下文中图形对象的部件表示模型和集成分类、识别与检测方法。通过深入研究和大量实验，项目组针对上述问题设计实现了相应的有效算法和框架，并在相关扩展领域如视频媒体中的图形对象处理、多媒体数据分析与检索等取得了若干创新方法。项目获得的成果，一方面发展了相应图形、文档识别与分析研究领域的方法与技术，尤其是在相对困难且较少研究关注和进展的复杂图形上下文数据处理技术方面，对其发展以及相关领域问题的研究起到了积极的推动作用，具有相当重要的学术价值。另一方面，项目研究获得的一系列方法对大量实际图形数据应用具有重要的参考价值，在后续工作中将重视其推广与应用。

中文关键词： 图形识别; 特征描述; 文档分析

英文摘要： To analyze, recognize and retrieve useful information from complex graphical data is an important aspect of pattern recognition and computer vision research, and also constitutes the key supporting techniques for the automatic and intelligent applications of graphical information. This research focuses on the effective feature description and representation techniques for objects in the complicated graphical contexts. We respectively explored multilevel, multiscale, various combinations of statistical and structural graphical features, which are effective and robust to various distortions and interferences imposed by the complicated contexts. Based on them, we further explored mechanisms of the part-based object model and ensemble classification, recognition and detection methods. Through intensive research and substantial experiments, we achieved corresponding effective algorithms or frameworks to these problems, and further extended our research to related areas like graphical object processing in video media and multimedia data analysis and retrieval, and developed several innovative methods. The results of the project research, on the one hand, developed methods and techniques in the graphic and document recognition and analysis research fields, especially for processing techniques for complicated graphical contexts that are relatively difficult and less addressed in research, which also contributed to the development of methods in related technical fields. On the other hand, the techniques acquired in this research have important reference value to large amounts of practical graphical applications, which will be put more emphasis in subsequent work.

英文关键词： graphic recognition; feature representation; document analysis