基于视觉语义主题与反馈日志的图像自动标注  

Image automatic annotation based on the visual semantic topics and feedback log

作　　者：孙君顶[1] 李海华[1] 靳姣林 

机构地区：[1]河南理工大学计算机科学与技术学院,焦作河南454000

出　　处：《光电子．激光》2017年第4期441-450,共10页Journal of Optoelectronics·Laser

基　　金：河南省基础与前沿技术研究(132300410462;112300410281)资助项目

摘　　要：为了提高图像标注性能,提出了一种基于视觉语义主题与反馈日志的图像自动标注方法。首先,提取图像前景与背景区域,分别进行处理;其次,基于WordNet构建标注词之间的语义关系模型,并结合概率潜在语义分析(PLSA)与高斯混合模型(GMM)建立图像底层特征、视觉语义主题与标注关键词间的联系,实现对图像的自动标注;然后,基于增量关联规则建立标注日志数据库,并在对数据库消噪的基础上,通过反馈技术提高标注的效果;最后,采用Corel5数据库进行验证,实验结果证明了本文方法的有效性。A novel automatic annotation scheme based on the visual semantic topics and feedback logs is proposed in the paper. Firstly, the foreground and background regions of the image are extracted and pro- cessed respectively. Then, the relations among the low-level features, the visual semantic topics and the key words are built based on the probabilistic latent semantic analysis （PLSA） and the Gaussian mixture model （GMM）. After that, the model of the semantic relations among the key words is constructed based on the WordNet to improve the annotation performance. Based on the incremental association rule, the relevance feedback technology is further introduced in the proposed annotation model by constructing the feedback logs database. The widely used database of Corel5 is used as the test bed,and the experimental results show that the new scheme gives better performance than the traditional methods.

关 键 词：图像标注 概率潜在语义分析(PLSA) 高斯混合模型(GMM) 反馈日志 增量关联规则 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]