跨媒体语义医学图像检索中关键技术研究

项目名称： 跨媒体语义医学图像检索中关键技术研究

项目编号： No.60873185

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 武器工业

项目作者： 吴洪

作者单位： 电子科技大学

项目金额： 30万元

中文摘要： 数字医学图像的数量正迅速增加，而当前落后的图像检索技术限制了对大量医学图像的有效利用。本项目探索通用医学图像检索中的关键方法和技术，主要围绕着跨媒体语义医学图像检索展开研究。"跨媒体"体现在互补地利用图像和文本信息提供多种检索方式，"语义"体现在通过图像自动分类技术自下而上地缩小图像特征表示与人类视觉认知间的"语义间隔"，通过基于概念的索引自上而下地缩小语义间隔，通过相关反馈技术动态地缩小这一间隔。主要研究内容包括：1）跨媒体语义医学图像检索框架2）结合特征选择的层次图像分类 3）基于概念的索引4）结合图像和文本信息的检索机制 5）相关反馈技术。通过在这些研究内容上的探索和创新，开发出高效的医学图像检索技术，为医学图像资源的充分利用，医疗服务质量和效率的提高打下坚实的技术基础。

中文关键词： 医学图像检索;特征选择;图像分类;相关反馈;基于概念的索引

英文摘要： The amount of digital medical images is growing rapidly in recent years, but the limited performance of current image retrieval techniques forms a major obstacle for efficient and effective use of these data. This project aims to develop and study the key methods and techniques for cross-media semantic retrieval of general medical images. "Cross-media" refers to combine visual and textual information to get better retrieval results than with each alone, and also to support different kinds of query formulations. "Semantic" means to retrieve images based on high-level semantics of images. And three approaches are taken together to solve the problem of semantic gap. Hierarchical image classification bridges the gap from the bottom up; concept-based indexing narrows the gap from the top down; and relevance feedback techniques reduce it dynamically. This project mainly includes five research topics: 1) a framework for cross-media semantic medical image retrieval, 2) hierarchical image classification with feature selection, 3) concept-based indexing, 4) retrieval mechanism combining image and textual features, 5) relevance feedback technique. With the investigations in these research topics, we expect to propose better medical image retrieval techniques to facilitate the use of digital medical images and lay foundation for real medical applications in the futures.

英文关键词： Medical Image Retrieval; Feature Selection; Image Classification; Relevance Feedback; Concept-Based Indexing