基于近义词自适应软分配和卡方模型的图像目标分类方法  

Image Object Classification Method with Homoionym Based Adaptive Soft-Assignment and Chi-Square Model

作　　者：赵永威 周苑[2] 李弼程[3] 柯圣财 

机构地区：[1]武警工程大学电子技术系,陕西西安710000 [2]河南工程学院计算机学院,河南郑州451191 [3]解放军信息工程大学信息系统工程学院,河南郑州450002

出　　处：《电子学报》2016年第9期2181-2188,共8页Acta Electronica Sinica

基　　金：国家自然科学基金(No.60872142;No.61379152);全军军事学研究生课题资助项目(No.YJS1062)

摘　　要：传统的视觉词典模型(Bag of Visual Words Model,Bo VWM)中广泛存在视觉单词同义性和歧义性问题.且视觉词典中的一些噪声单词—"视觉停用词",也会降低视觉词典的语义分辨能力.针对这些问题,本文提出了基于近义词自适应软分配和卡方模型的图像目标分类方法.首先,该方法利用概率潜在语义分析模型(Probabilistic Latent Semantic Analysis,PLSA)分析图像中视觉单词的语义共生概率,挖掘图像隐藏的语义主题,进而得到语义主题在某一视觉单词上的概率分布;其次,引入K-L散度度量视觉单词间的语义相关性,获取语义相关的近义词;然后,结合自适应软分配策略实现SIFT特征点与若干语义相关的近义词之间的软映射;最后,利用卡方模型滤除"视觉停用词",重构视觉词汇分布直方图,并采用SVM分类器完成目标分类.实验结果表明,新方法能够有效克服视觉单词同义性和歧义性问题带来的不利影响,增强视觉词典的语义分辨能力,较好地改善了目标分类性能.The synonymy and ambiguity of visual words always exist in the conventional bag of visual words model based object classification methods. Besides,the noisy visual words,so-called＂visual stop-words＂will degrade the semantic resolution of visual dictionary. In this article,an image object classification method with homoionym based adaptive soft-assignment and chi-square model is proposed to solve these problems. Firstly,PLSA（ Probabilistic Latent Semantic Analysis）is used to analyze the semantic co-occurrence probability of visual words,excavate the latent semantic topics in images,and get the latent topic distributions induced by the words; Secondly,the KL divergence is adopted for measuring semantic distance between visual words,which can get semantically related homoionym; then,adaptive soft-assignment is proposed to realize the soft mapping between SIFT features and some homoionym; finally,the Chi-square model is introduced to eliminate the＂visual stop-words＂and reconstruct the visual vocabulary histograms,and moreover,SVM（ Support Vector Machine） is used to accomplish object classification. Experimental results indicated that the adverse effects produced by the synonymy and ambiguity of visual words can be overcome effectively,the distinguishability of visual semantic resolution is improved,and the image classification performance is substantially boosted compared with the traditional methods.

关 键 词：视觉词典模型 概率潜在语义分析模型 K-L散度 卡方模型 目标分类 

分 类 号：N391[自然科学总论]