一种基于GMM的图像语义标注方法  

GMM-Based Image Semantic Annotation Method

作　　者：陈晓[1] 张尤赛[1] 邹维辰[1] 

机构地区：[1]江苏科技大学电子信息学院,江苏镇江212003

出　　处：《电视技术》2012年第23期35-38,共4页Video Engineering

基　　金：江苏高校优势学科建设工程

摘　　要：针对图像语义概念具体语义描述的问题,提出了一种基于GMM的图像语义标注方法。该方法对于每一个语义概念分别建立基于颜色特征和纹理特征的GMM模型,利用EM算法获取关键词内容,最后融合两个GMM模型求取的概率排序结果,对未知图像进行标注。实验结果表明,提出的方法能够准确地为待标注的图像预测出若干文本关键字,有效提高图像标注的查准率和查全率。A GMM-based image annotation method is proposed in this paper for the specific description of the semantic concept. GMM models based on color feature and texture feature for each semantic concept are respectively built. And then, EM algorithm is used to obtain the content of keywords. E- ventually, syncretized the sorting results of probability by the two GMM models to annotate the unknown images. Experimental results show that the meth- od can predict text keywords for the unannotated images more accurately and have valid precision and recall.

关 键 词：GMM 语义概念 EM算法 图像标注 

分 类 号：TN911.73[电子电信—通信与信息系统]