相关反馈修正语义流形的图像检索方法  

Image Retrieval Method of Relevance Feedback Revise Semantic Manifold

作　　者：刘利[1] 

机构地区：[1]惠州学院计算机科学系,广东惠州516007

出　　处：《小型微型计算机系统》2014年第12期2759-2763,共5页Journal of Chinese Computer Systems

基　　金：惠州市科技计划项目(2012-7;2012-5)资助

摘　　要：目前结合相关反馈和流形学习的图像检索方法,都是直接利用相关反馈中的类别信息,并未结合可视化信息推论出隐藏的有价值信息.本文对增强关系嵌入算法进行扩展,基于该方法构建可视化关系图,并结合用户相关反馈信息中的不同类别图像之间的关系推论学习,据此去除低层可视化关系图中的错误连接以修正语义流形,从而获得更准确的投影转换矩阵.实验结果表明本文方法将图像映射到低维的空间时可获得更高的检索精度.The image retrieval methods combined relevance feedback and manifold learning attracts a lot of researchers. However,these methods used category information from relevance feedback directly and have not combined visual information to infer hidden valuable information. This paper modify Augmented Relation Embedding algorithm. Firstly, a visual relation graph is constructed based on Augmented Relation Embedding algorithm. Secondly, the relation of images in different category from users＂ relevance feedback is combined with visual relation graph to infer learning. Finally, inference information is used to remove the incorrect connection in the visual relation graph for revising semantic manifold, so that more accurate transformation matrix of projection is gained. Experimental results show that this method can improve retrieval precision while map images to a lower dimensional space.

关 键 词：流形学习 基于内容的图像检索 相关反馈 修正语义流形 推论学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]