语义检索和自动识别的唐卡图像知识库系统关键技术研究

项目名称： 语义检索和自动识别的唐卡图像知识库系统关键技术研究

项目编号： No.60875006

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 轻工业、手工业

项目作者： 王维兰

作者单位： 西北民族大学

项目金额： 30万元

中文摘要： 藏族唐卡具有鲜明的文化特点，浓厚的宗教色彩和独特的艺术风格，堪称藏族的百科全书，是我国重要的文化遗产。为研究唐卡图像视觉特征、领域知识、语义特征提取与表示等，完成一个具有一定查准率和查全率的唐卡图像检索知识库系统，主要内容包括唐卡图像视觉特征、领域知识和内容相结合的高层语义提取和表示；唐卡图像领域知识模型的构建，通过图像中不同对象之间关联关系的推理达到图像分类的方法。重要的结果：通过对唐卡图像资源的收集、梳理、归纳和唐卡图像知识库的研究与构建，应用各种视觉特征提取方法、根据唐卡图像特点的分块特征提取以及特征融合等实证研究，获得了具有一定查准率和查全率的基于内容的唐卡图像检索方法；提出了唐卡图像概念标注方法、以图学习自动标注算法为框架的分类器、描述概念和语义关键词以及语义关键词之间模糊关系的C-K关系网，能够很好的支持唐卡图像高层语义检索；此外提出了唐卡图像与非唐卡图像的区分方法、结合破损块形状和邻域分类的唐卡图像修复方法等。将唐卡元数据、原始像元信息、视觉特征、图像对象、领域知识、藏汉英三种文字的内容描述等信息纳入统一的分析框架，为视觉分析和领域知识指导下的图像检索奠定了良好的基础。

中文关键词： 唐卡图像；领域知识；视觉特征；图像检索

英文摘要： Thangka art is an important heritage accumulated in the social, historical and cultural development for thousand years, and has always been regarded as treasures by Tibetan people, and they are regarded as an encyclopedia of the history and society. In order to study features extraction and representation of visual features, domain knowledge and semantic, to complete a Thangka image retrieval knowledge base system with certain precision and recall, the primary content include features extraction and representation of visual features, domain knowledge and high level semantic features of Thangka image; construction model of domain knowledge of Thangka image,and classification method through inference of relationship between objects in image. Important results: through to the resource collection, analysis, induction and construction on knowledge base of Thangka image, apply various visual feature extraction methods, and made many empirical studies according to the characteristics of Thangka image block features extraction and feature fusion,and get with a certain precision and recall of content-based image retrieval method; the concept annotation method of Thangka imag, and based on the framework of graph learning automatic annotation algorithm are proposed, and construct the C-K (concept semantic keywords) relation net to be used to describe the concept and semantic keywords and fuzzy relation between semantic keywords, so that can well support Thangka image high-level semantic retrieval; In addition Thangka image and non-Thangka image distinguishing method and image restoration method combining the shape of damaged patches and type of neighborhood patches have been developed.The Thangka image metadata, the original pixel information, visual features, image objects, domain knowledge, Tibetan-Chinese-English three text description of the contents of information into a unified framework, for the visual analysis and domain knowledge under the guidance of image retrieval has laid a good foundation.

英文关键词： Thangka image; domain knowledge; Vision feature; Image retrieval