项目名称: 大规模图像近似拷贝快速检测方法研究
项目编号: No.61202280
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 计算机科学学科
项目作者: 姚金良
作者单位: 杭州电子科技大学
项目金额: 24万元
中文摘要: 图像近似拷贝检测是图像检索领域一个新的研究方向。在大规模图像库中实现快速鲁棒的图像近似拷贝检测,尚需解决图像的紧致描述与高效检索等关键问题。当前基于视觉词汇的方法对视觉词汇的多义性和同义性问题缺少针对性的关注。本项目拟以视觉词汇的多义性和同义性问题为出发点,研究面向大规模图像库的有效利用视觉词汇上下文信息和视觉词汇扩展的图像近似拷贝快速检测方法。研究内容包括:(1)局部特征视觉词汇化方法,使视觉词汇具有紧致性、快速映射、强区分能力、数量有限性;(2)视觉词汇上下文建模方法;(3)大规模高维视觉词汇的快速近邻检索;(4)高效的图像近似拷贝验证方法。通过本项目的研究,有效弥补当前在大规模图像库中视觉词汇表示和检索方面的不足,推动近似拷贝图像检测技术的进一步发展,相关研究成果有助于促进基于视觉词汇的图像分类和检索研究。
中文关键词: 图像检索;近似拷贝图像检索;场景检索;视觉词汇;上下文描述子
英文摘要: The image near-duplicate detection is a new research direction in image retrieval domain. There are two key problems, image compactness description and fast retrieval, required to be solved to implement fast and robust detection of near-duplicate images in a large-scale image database. The polysemy and synonymity of visual words get less attention from the existing methods based on visual words. In this project, we will research fast detection methods of near-duplicate images by making use of visual words context and visual words extending for a large-scale image database. The main research points include: (1)The transformation method from local features to visual words which are compact, fast mapping, strongly discriminating and limited count; (2) The modeling method of visual words context; (3) The high-dimensional visual words fast retrieval method in a large-scale visual words database; (4) The efficient verification method of near-duplicate images. The research work in this project can improve the description and retrieval efficiency of visual words, and promote detection technology of near-duplicate images. The achievements can also contribute to the research work on visual words-based image classification and retrieval.
英文关键词: Image Retrieval;Near-duoplicate Image Retrieval;Scene Image Retrieval;Visual Word;Contextual Descriptor