基于几何不变量和规格化直方图的层次图像检索模式  

Novel hierarchical image retrieval using geometric invariants and normalized histogram

作　　者：刘继光[1] 王建军[2] 张明新[2,3] 郑金龙[2,3] 

机构地区：[1]甘肃省广播电影电视总台,兰州730000 [2]常熟理工学院计算机科学与工程学院,江苏常熟215500 [3]西北师范大学数学与信息科学学院,兰州730070

出　　处：《兰州大学学报（自然科学版）》2010年第1期129-132,共4页Journal of Lanzhou University(Natural Sciences)

基　　金：甘肃省自然科学基金项目(3ZS051-A25-047)

摘　　要：提出层次图像分割方法,将检索图像在不同的层次上分割为不同大小的子块,计算这些子块与查询图像间的相似度进而实现检索,同时可以获得检索图像中查询目标对象的尺度和位置信息。根据图像的几何不变量和规格化直方图特征提出了一种层次块图像检索模式,并且用500幅各类图像进行了试验。在最理想的相似度阈值下,仅以几何不变量为特征的检索精度可达到78%。此外,该检索模式也可很好地处理大小不同的图像。Hierarchical image segmentation was proposed to partition the retrieved images into equal blocks with different sizes at difficult levels. Then, the similar metrics of these sub-blocks to query image were evaluated to retrieve those sub-blocks with contents in query images. Meanwhile, information about scales and locations of query objects in retrieved images could also be retrieved. The hierarchical block image retrieval schemes using geometric invariants and normalized histograms were tested by experiments via a database with 500 images. The retrieval accuracy with geometric invariants as invariant features could achieve 78% for the optimal similar metrics threshold. Furthermore, the schemes could also work with different size images.

关 键 词：图像检索 层次分割 几何不变量 规格化直方图 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]