一种基于上下文语义信息的图像块视觉单词生成算法  

A Novel Visual Words Definition Algorithm of Image Patch Based on Contextual Semantic Information

作　　者：刘硕研[1] 须德[1] 冯松鹤[1] 刘镝[2] 裘正定[2] 

机构地区：[1]北京交通大学计算机科学系,北京100044 [2]北京交通大学信息科学研究所,北京100044

出　　处：《电子学报》2010年第5期1156-1161,共6页Acta Electronica Sinica

基　　金：国家自然科学基金(No.60803072;No.90820013);中国博士后科学基金(No.20090460197)

摘　　要：基于视觉单词的词包模型表示(Bag-of-Words)算法是目前场景分类中的主流方法.传统的视觉单词是通过无监督聚类图像块的特征向量得到的.针对传统视觉单词生成算法中没有考虑任何语义信息的缺点,本论文提出一种基于上下文语义信息的图像块视觉单词生成算法:首先,本文中使用的上下文语义信息是视觉单词之间的语义共生概率,它是由概率潜在语义分析模型(probabilistic Latent Semantic Analysis)自动分析得到,无需任何人工标注.其次,我们引入Markov随机场理论中类别标记的伪似然度近似的策略,将图像块在特征域的相似性同空间域的上下文语义共生关系有机地结合起来,从而更准确地为图像块定义视觉单词.最后统计视觉单词的出现频率作为图像的场景表示,利用支持向量机分类器完成图像的场景分类任务.实验结果表明,本算法能有效地提高视觉单词的语义准确性,并在此基础上改善场景分类的性能.Visual Words （VWs） representation has recently become popular for scene classification. Visual words are usually constructed by using an unsupervised method to cluster the appearance descriptor vectors of patches. However, the traditional visual words method considers nothing about semantic information. To overcome this defect,this paper proposes the novel contextual semantic constrained Markov Random Fields （MRF） visual words. There are two areas of novelty：first,the useful contextual semantic constrained information are the semantic co-occurrence probabilities of image patches obtained by probabilistic Latent Semantic Analysis （pLSA） instead of by manually labeled. And then we introduce the pseudo-likelihood of labeling to combine the feature appearance similarity and contextual semantic information together in order to provide the more accurate visual words representation. Once these visual words are obtained, the frequency of these visual words in an image forms a histogram which is subsequently used in a scene categorization task by the Support Vector Machine （SVM） classifier. The experimental results compared with the existing methods show that contcxtual semantic constraints indeed do benefit for visual words selection.

关 键 词：场景分类 视觉单词 概率潜在语义分析模型 MARKOV随机场模型 上下文语义信息 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]