基于颜色和形状特征的彩色图像检索方法  

Color Image Retrieval Based on Color and Shape Features

作　　者：孙君顶[1] 崔江涛[1] 毋小省[2] 周利华[1] 

机构地区：[1]西安电子科技大学多媒体研究所,西安710071 [2]焦作大学计算机系,河南焦作454000

出　　处：《中国图象图形学报（A辑）》2004年第7期820-827,共8页Journal of Image and Graphics

基　　金：十五国际科技 (电子 )预研项目(4 13 160 5 0 1)

摘　　要：针对基于内容的彩色图像相似性检索问题 ,提出了一种新的相似性彩色图像检索方法 ,该方法既考虑了图像的颜色特征 ,也考虑了图像的空间信息特征 ,即在对HSV颜色模型进行特殊处理的基础上 ,将提取的色调不变量作为图像的颜色特征 ,同时设计了图像状态矩阵来描述图像的形状信息和空间位置信息。在进行图像间的相似性测量时 ,为了结合不同的子特征进行全局的相似性检索 ,还采用Guassian模型对不同子特征间的距离进行了归一化处理。用不同类型的图像对这种方法进行的试验结果表明 ,它用于图像的相似性检索是很有效的 。The color histogram based image retrieval method is simple and invariant for translation and rotation of the images but losing the spatial information of the color. Recently many methods, such as accumulative histogram, color correlograms, local color histogram, etc, are introduced to improve the color histogram method. In this paper, a new content\|based color image retrieval method is proposed, in which both the color content and the shape feature of the image have been taken into account. Firstly, based on the special disposal on the HSV color space, an improved accumulative histogram of the hue is calculated as the color feature. To attain the spatial information, H-, S-, and V-component of the image are firstly divided into n×n blocks which are classified into 3 status, flatness, texture and edge status. Then each gray image is translated into a matrix composed of those 3 status values. After that the status matrix is transformed into 1\|dimension status sequence, the transition probability matrix of the sequence is calculated as the image's spatial distribution information. In matching the similarity of the images, the Guassian model is used to normalize the different sub\|characters distance. Experiments with different kinds of images indicate that this method is great effective in image's retrieval.

关 键 词：相似性检索 色调直方图 状态矩阵 图像检索 转移概率矩阵 HSV颜色模型 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]