基于多种高级语义特征的图像检索技术  

A new semantic-based color image retrieval

作　　者：王向阳[1,2] 于永健[1] 杨红颖[1] 周璐[1] 

机构地区：[1]辽宁师范大学计算机与信息技术学院,辽宁大连116029 [2]北京大学视觉与听觉信息处理国家重点实验室,北京100871

出　　处：《辽宁师范大学学报（自然科学版）》2009年第4期445-452,共8页Journal of Liaoning Normal University:Natural Science Edition

基　　金：国家自然科学基金资助项目(60773031;60873222);视觉与听觉信息处理国家重点实验室(北京大学)开放基金资助项目(0503);计算机软件新技术国家重点实验室(南京大学)开放基金资助项目(A200702)

摘　　要：基于语义内容的图像检索已成为解决图像低层特征与人类高级语义之间"语义鸿沟"的关键.笔者以性能优越的回归型支持向量机(SVR)理论为基础,结合重要的图像边缘信息及人眼视觉特性,提出了一种基于多种高级语义特征的图像检索新方法.该方法首先利用Canny检测算子提取原始图像的边缘信息,并据此得到低层纹理特征与颜色特征,同时利用SVR将低层特征映射到高级语义,以获得图像的高级对象语义.然后结合图像边缘线条方向,利用SVR将线条方向映射为高级语义,以获得图像的线条情感语义.再结合人眼视觉系统感知特性,给出基于全局主要颜色的高级颜色语义.最后根据上述多种高级语义特征进行图像检索.实验结果表明,该方法能够有效地对图像高级语义进行刻画,不仅图像匹配检索效果良好,而且具有稳定的检索性能,其对于缩小低层视觉特征与高级语义概念之间的"语义鸿沟"具有重要意义.The performance of content-based image retrieval（CBIR） systems is largely limited by the gap between the low-level feature and high-level semantic concept.In this paper,a new semantic-based color image retrieval scheme is proposed,which not only takes into consideration the important image edge information and human visual system,but also utilizes the support vector regression（SVR） theory.Firstly,the important image edge is extracted by using canny detection operator and the low-level texture feature and color feature are computed.Then,the high-level object semantic is determined by mapping the low-level feature using SVR.Secondly,the high-level emotion semantic is determined by mapping the low-level line direction using SVR.Thirdly,the high-level color semantic can be obtained by using the main color of the quantized color image.Finally,image retrieval is implemented by using the above high-level object semantic,emotion semantic and color semantic.Experimental results show that the proposed image retrieval are effective in characterizing image high-level semantic and can provide sound and robust image retrieval performance,which has strong significance for reducing the ＂semantic gap＂ between the visual feature and semantic concept.

关 键 词：图像检索 高级语义 SVR 对象语义 情感语义 颜色语义 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]