项目名称: 面向语义概念和上下文关系的图像检索关键技术研究
项目编号: No.60875011
项目类型: 面上项目
立项/批准年度: 2009
项目学科: 无线电电子学、电信技术
项目作者: 杨育彬
作者单位: 南京大学
项目金额: 29万元
中文摘要: 本项目围绕面向语义概念和上下文关系的图像检索中的若干关键技术及其应用开展研究。以图像的语义概念处理技术为核心,在深入研究图像的颜色、形状、纹理、空间分布、空间拓扑关系等多模态的底层视觉特征选择和融合的基础上,运用模式识别、人工智能、机器学习以及语义网等技术,引入形式化的知识本体表示,建立图像的上下文统计模型;研究图像的语义概念学习和标注算法,建立面向语义概念和上下文关系的、多层次、具有互操作性、融合全局及局部信息的图像语义内容描述模型及检索机制;此外,将开展对可计算的主观语义概念认知理论的探索研究,提出结合概率学习模型与知识本体推理的复杂语义概念推理方法和结合知识本体的语义概念主观相似性计算模型,并基于用户检索上下文的统计模型进行语义相关反馈算法和个性化检索研究,最终实现一个基于Internet 的分布式图像语义检索原型系统,以更好地实现对图像的语义概念的提取、索引和检索。
中文关键词: 语义概念;图像检索;机器学习;上下文关系;知识本体
英文摘要: The project studies on the key techniques and applications of image retrieval based on semantic concepts and contexts, which emphasizes on designing and implementing semantic concepts processing methods. The project firstly conducts in-depth research on selection and fusion of multimodal visual features including colors, shapes, textures, spatial relationships and topological structures, based on which a novel statistical model for describing image contexts is established by combining pattern recognition, artificial intelligence, machine learning, semantic web and ontology theories. Then, the project investigates learning and annotating methods for semantic concepts, and establish a description model and a retrieval paradigm for representing and sensing image semantic concepts. Moreover, the project also explores on a computational theory for subjective semantic perception and proposes a subjective similarity mesurement for semantic concepts, together with a reasoning mechanism for complex semantic concepts combining statitiscal learning techniques and ontology-based reasoning methods. After that, the project studies on semantic-based user relevance feedback algorithms based on the statistical context model built from the users' retrieval records, and finally implements an Internet-based, distributed prototype systems for concept-based image retrieval system.
英文关键词: semantic concepts;image retrieval; machine learning; context relationship; ontology