基于综合特征的彩色图像检索算法  

Color Image Retrieval Based on Multiple Features

作　　者：高燕[1] 胡学龙[2] 阮文佳[3] 李玲[1] 丁宁[1] 

机构地区：[1]南京信息职业技术学院电子信息学院,江苏南京210046 [2]扬州大学信息工程学院,江苏扬州225009 [3]中国海洋大学信息科学与工程学院,山东青岛266100

出　　处：《江南大学学报（自然科学版）》2009年第6期670-673,共4页Joural of Jiangnan University (Natural Science Edition)　

基　　金：江苏省图像处理与图像通信重点实验室资助项目(ZK206008)

摘　　要：针对单一特征不足以准确描述图像的问题,提出了一种综合颜色和形状特征的图像检索算法。该算法对HSV颜色空间进行特殊处理并量化编码,提取各像素点及其多邻域颜色矩的归一化直方图作为颜色特征;对图像进行整数小波变换后的低频分量分块并编码,提取一步转移概率矩阵作为形状特征;综合利用颜色、形状两个特征计算图像间的相似度,并进行彩色图像检索。仿真实验结果表明,该算法能够准确和高效地查找出用户所需内容的彩色图像,且具有较好的查准率和查全率。In the paper, a new method of content-based image retrieval is presented, which integrates color and shape features. Firstly, the color image is quantized and encoded based on the special disposal on the HSV color space and the normalized histograms are extracted from every pixel and its multi-neighbor color moments as color feature. Secondly, low frequency band of integer wavelet transform is segmented into blocks and encoded and the transition probability matrix is calculated as shape information. Finally, the similarities between color images are eomputed by using a combined index based on color feature and shape feature. The experimental results show that the proposed algorithm is more accurate and efficient in retrieving the user-interested images.

关 键 词：图像检索 整数小波变换 多邻域颜色矩直方图 一步转移概率矩阵 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]