基于视觉内容语义相关度的图像标签自动排序方法  

Automatic image tag ranking scheme based on visual content semantic relatedness

作　　者：赵英海[1] 查正军[2] 李珊珊[1] 吴秀清[1] 

机构地区：[1]中国科学技术大学电子工程与信息科学系,安徽合肥230027 [2]新加坡国立大学计算机学院,新加坡639798

出　　处：《中国科学技术大学学报》2011年第2期108-115,共8页JUSTC

摘　　要：提出一种基于视觉内容语义相关度的图像标签自动排序方法.该方法按照标签与图像内容的语义相关程度对网络共享图像的标签进行排序.首先,算法基于贝叶斯理论给出标签与图像内容语义相关度计算的概率表述.然后,融合多种视觉特征以实现对不同语义的标签与图像内容相关度概率的准确估计,具有较高的可扩展性.实验数据采用149 915幅Flickr网站下载图像,实验结果验证了本文方法的有效性.An automatic image tag ranking scheme was proposed based on visual content semantic relatedness, which sorted the tags of the community-contributed images according to the semantic relatedness between tags and image contents. Firstly, the semantic relatedness calculation between tag and image content was formulated as a probabilistic problem based on Bayes＇ theorem. Then, multiple visual features were fused to obtain reasonable probability evaluation for tag-image content relatedness in different semantic themes. This method has high scalability. Extensive experiments were conducted over a dataset consisting of 149 915 images downloaded from Flickr and experimental results demonstrate the effectiveness of the proposed method.

关 键 词：FLICKR 标签排序 内容相关度 视觉内容 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]