图像纹理特征相似性度量的研究  

Research on the Similarity Measurement of Image Texture Features

作　　者：黄丽达[1] 邹北骥[2] 

机构地区：[1]湖南大学计算机与通信学院,湖南长沙410082 [2]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《计算机工程与科学》2006年第7期49-52,共4页Computer Engineering & Science

基　　金：湖南省自然科学基金资助项目(02JJY3049)

摘　　要：纹理是图像的重要属性,基于纹理特征检索图像是当前的研究热点,对图像的纹理进行相似性比较是进行图像检索的关键。根据纹理的特点,本文将通用的向量空间模型进行拓展,构建了一个针对簇集进行相似性匹配的模型———聚类空间模型,对图像纹理相似性进行度量,并据此实现了无需分割的多纹理图像检索。我们分别针对单纹理图像和自然图像库进行了实验,获得的实验结果与人类视觉认知的结果一致。Texture is an important item of image information, texture-based image retrieval has been an active research area, and the similarity comparison of texture features is a key to image retrieval. In this paper, based on the properties of texture, by expanding the Vector Space Model （VSM）, we create a new model for matching the similarities of feature signatures to measure the similarities of texture images, which is called Clusters Space Model （CSM）. And based upon it, a new method of multiple texture image retrieval without segmentation is shown. By comparing with other similarity measures for texture features, experimental results indicate it is consistent with human perception.

关 键 词：图像纹理 图像检索 聚类空间模型 相似性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]