摘要
基于内容的图像检索的关键问题之一是高层语义和低层图像特征之间的差异,相关反馈技术是缩短这个"语义鸿沟"的有效方法。本文提出了一种新的相关反馈算法,通过分析正例图像在特征空间中的散布来构造该类图像的投影空间,该空间对应于一个语义类在特征空间中分布密集的子空间,在投影空间中计算相似图像。同时根据每次反馈的信息不断修正投影空间来提高系统的检索性能。在Corel大图像库中的实验结果表明,该算法对多例图像查询有较好的检索效果。
One of the key issues in content-based image retrieval is the disparity between high-level semantic concepts and low-level image features. Relevance feedback is an effective technique to bridge “the great semantic gap”. A novel relevance feedback approach is proposed in this paper. By analyzing the scatter of the positive images in the feature space, similar images are calculated in a projected space, corresponding to a subspace of the feature space where the images belonging to the semantic group distribute more closely. Moreover, the projected space is adjusted to the informa- tion of each round of feedback, thus improving the system's retrieval performance. Experimental results on the large Corel image collection show that this algorithm achieves a good retrieval performance in querying multi images.
出处
《计算机科学》
CSCD
北大核心
2005年第12期213-215,231,共4页
Computer Science
基金
国家973计划项目(G1998030500)
关键词
基于内容的图像检索
相关反馈
多例查询
Content-based image retrieval (CBIR), Relevance feedback, Multi-image query