一种基于兴趣点的图像检索方法  

A Better Image Retrieval Approach

作　　者：鲍永生[1] 韩军伟[1] 郭雷[1] 

机构地区：[1]西北工业大学自动化学院,陕西西安710072

出　　处：《西北工业大学学报》2004年第2期192-195,共4页Journal of Northwestern Polytechnical University

基　　金：国家自然科学基金 (6 0 175 0 0 1);973项目

摘　　要：提出一种基于兴趣点的图像检索方法。此方法主要有兴趣点检测、基于兴趣点的特征描述和相似度度量 3个步骤。先使用一个自适应滤波器平滑图像 ,然后提取兴趣点 ;设计一个包含兴趣点局部灰度变化、兴趣点相互位置关系和兴趣点分布的 3D直方图来表征图像特征 ;用图像间 3D直方图距离来测量图像间的相似程度。在含有 32 0 0幅图像的数据库上的大量实验结果证明 :本文提出的算法是有效的。Refs.7 and 8 offered image retrieval approaches that are, in our opinion, not quite efficient. Like Ref.7, we use interest points and like Ref.8, we use anisotropic diffusion. Unlike Ref.7, we design a self-adaptive filter that can significantly lower the error of detection of interest points; subsection 1.2 gives in some detail our design of this self-adaptive filter. Unlike Ref.8, we do not treat interest points as being isolated; instead we attach great importance to essential features about interest points as a whole; section 2 gives in some detail our consideration of these features. A fair amount of test results, based on a database of 3200 images and shown in Figs.8 and 9, show preliminarily that our image retrieval approach is indeed better than Refs.7 and 8 in that our approach can significantly raise the accuracy of error detection and decrease the time of detection.

关 键 词：图像检索 兴趣点 自适应平滑滤波器 3D直方图 

分 类 号：TN911.73[电子电信—通信与信息系统]