项目名称: Web图像视觉模式挖掘及其应用
项目编号: No.61201446
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 电子学与信息系统
项目作者: 黄俊
作者单位: 中国科学院上海高等研究院
项目金额: 24万元
中文摘要: 图像挖掘技术致力于对海量图像数据自动分析处理,以获取有意义的模式和知识。本课题围绕Web图像的视觉模式发现和提取展开,重点研究了基于显著性和基于语义的视觉模式挖掘方法,并将这两种方法融合用于提升图像检索的重排序性能。通过研究视觉选择性注意机制的计算模型,拟从统计学习的角度出发,建立多尺度视觉显著性模型。研究基于语义的视觉模式挖掘问题,设计无监督的分析方法对来自于Web的图像自动进行聚类处理,从而提取出特定语义概念的重复视觉模式。本课题还将研究融合两种视觉模式的图像重排序算法,将视觉上显著的且与查询主题密切相关的图像优先返回给用户。本课题的实施将扩展Web图像视觉模式挖掘的研究范畴,利用所提出的挖掘算法,不仅能用于提升图像浏览和检索系统的性能,同时也能在图像分类、图像标注等方面得到实际应用。
中文关键词: 图像挖掘;图像识别;视觉显著性;图像检索;用户意图
英文摘要: Image mining technology has been applied in automatic analyzing and processing large scale image data for the purpose of obtaining meaningful patterns and knowledge. Focusing on Web image visual pattern discovery and extraction, the project mainly investigates the methods of saliency-based and concept-based visual pattern mining, and attempts to integrate the two methods to improve the performance of image re-ranking. The project probes into the computational model of visual selective attention mechanism. We plan to build a multi-scale visual saliency model based on statistical learning. Furthermore, we will design the unsupervised method of automatic clustering for Web images, and extract the recurrent visual pattern of specific concept. This project will also attempt to work out a image re-ranking algorithm integrating the two visual patterns, by which the images which are visually salient and closely related to the search query will have high ranks. The implementation of project will expand the research category of Web image pattern mining. This proposed visual pattern can not only be used to improve the performance of image browsing and retrieval, but can also be applied in some other fields such as image classification and image annotation.
英文关键词: Image Mining;Image Recognition;Visual Saliency;Image Retrieval;User Intention