融合LDA和多类SVM的图像语义映射研究  

Research on image semantic mapping combining with multi-class SVM and LDA

作　　者：赵炜[1] 陈俊杰[1] 李海芳[1] 

机构地区：[1]太原理工大学计算机与软件学院,太原030024

出　　处：《计算机工程与应用》2009年第18期164-166,共3页Computer Engineering and Applications

基　　金：国家自然科学基金No.60773004;山西省自然科学基金No.2007011050~~

摘　　要：建立图像低层特征到高层语义的映射是图像语义检索的关键问题之一,SVM是其中行之有效的方法。为了便于规则生成,将模糊C均值聚类SVM多类分类方法应用于图像语义映射。但由于异类图像特征常常混杂,最终形成的二叉树分支一般很多,映射准确率下降明显。为此,将线性判别分析法引入二叉树建树过程中,通过聚类之前先对特征优化处理来改进算法性能。实验结果表明该方法建立起了更便于理解的分类树结构且LDA的引入使得映射准确率有所提高,满足了图像语义映射的要求。Correlating image low-level feature with high-level semantic is one of the key questions of image semantic retrieval, SVM is an effective way.Multi-class SVM based on fuzzy C-means clustering is introduced to image semantic retrieval in order to facilitate the generation of rules.But because heterogeneous features of images are always motley, a great many branches of binary-tree are formed and mapping accuracy rate comes down obviously.Therefore,linear discriminant analysis is introduced to binary-tree to improve the algorithm＇s performance through pretreatment before clustering.The results show that the method builds up a more comprehensible tree configuration and improves the mapping correct rate due to LDA,and meets the require of image semantic mapping.

关 键 词：图像语义分类 支持向量机 模糊C均值聚类 线性判别分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]