摘要
目前,K-means聚类算法所构建的视觉词典已无法满足用户对图像分类的需求,为了提高视觉词汇码本的质量和图像分类的准确率,针对构建视觉词典的算法进行研究。在空间金字塔模型的基础上,图像分类算法首先采用K-means算法对SIFT特征进行初步聚类,得到一个粗略的划分,然后利用层次聚类进行精确归类,最后对视觉词典进行特征编码并且用SVM分类器进行分类;在混合的聚类算法中引入基于信息熵的属性加权方法,通过信息熵度量类间及类内的相似性。在Catchl01和Catch256图像库上的实验结果表明,与传统的Kmeans算法和加权K-means算法相比,结合信息熵的混合聚类算法能够有效提高空间金字塔模型的分类准确率。
At present,the visual dictionary constructed by K-means clustering algorithm can not meet the demand of image classification.In order to improve the quality of visual vocabulary codebook and the accuracy of image classification,this paper studied the algorithm of constructing visual dictionary.On the basis of the spatial pyramid model,the image classification algorithm first used the K-means algorithm to initialize the SIFT feature in order to get a rough division,and then it accurately classified by hierarchical clustering uses.Finally,it coded the formed visual dictionary by feature coding algorithm and classified by SVM classifier.In the hybrid clustering algorithm,it introduced the attribute weighting method based on information entropy,and measured the similarity between class and class by information entropy.The experimental results on Catchl01 and Catch256 image banks show that the hybrid clustering algorithm combined with information entropy can effectively improve the classification accuracy of spatial pyramid model compared with the traditional K-means algorithm and the weighted K-means algorithm.
作者
刘明波
胡朝举
Liu Mingbo;Hu Chaoju(School of Control&Computer Engineering,North China Electric Power University,Baoding Hebei 071000,China)
出处
《计算机应用研究》
CSCD
北大核心
2018年第11期3476-3478,共3页
Application Research of Computers
关键词
层次聚类
信息熵
空间金字塔模型
图像分类
K-MEANS聚类
hierarchical clustering
information entropy
spatial pyramid model
image classification
K-means clustering
