显著兴趣点颜色及空间分布熵的图像检索方法  

A Novel Image Retrieval Algorithm Based on Color and Distribution Entropy of Prominent Interest Points

作　　者：曾智勇[1] 安志勇[2] 周利华[2] 

机构地区：[1]福建师范大学软件学院,福建福州350007 [2]西安电子科技大学ISN国家重点实验室,陕西西安710071

出　　处：《红外技术》2007年第3期160-163,共4页Infrared Technology

基　　金：十五国防科技(电子)预研项目

摘　　要：提出了一种利用显著兴趣点进行图像检索的新方法,此方法主要有显著兴趣点检测、基于显著兴趣点的特征描述和相似性度量三个步骤。先使用一个自适应滤波器对图像进行滤波,然后提取显著兴趣点;以显著兴趣点为线索,设计了一种基于显著兴趣点的颜色分布熵,既利用了显著兴趣点的局部特征,又考虑了显著兴趣点的空间分布结构;用图像间的颜色分布熵来度量图像间的相似性。该检索算法不但保证了对图像旋转、平移鲁棒性,而且克服了传统直方图没有空间位置的缺陷。实验结果表明,该方法对图像检索非常有效。A A novel image retrieval algorithm based on prominent interest points is presented. The algorithm consists of three steps： prominent interest points detection, feature description based on prominent interest points and similarity measurement, image is filtered by a self-adaptive filter, then prominent interest points are extracted; prominent interest points are regarded as clues, and color distribution entropy is designed, which takes not only the local color feature into consideration, but also the space distribution information of prominent interest points; the similarities are measured by the distance of color distribution entropy. With robustness to rotated and translated image, the algorithm avoids shortcoming of losing the location information in color histogram. The results show that this algorithm is efficient in image retrieval.

关 键 词：图像检索 显著兴趣点 环形颜色直方图 颜色分布熵 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]