摘要
为了提高自动标注系统的性能,提出了一种基于最优标签集图像自动标注系统优化算法.用词袋模型表示图像,采用CTM模型进行图像标注,在此基础上,采用基于词频因子的词间相关性以及启发式迭代算法对获得的标注词进行有效的优化,提高了标注词的准确性.在Corel5K数据集中利用LDA模型和CTM模型进行图像标注对比实验,实验结果表明本文提出的图像标注方法能有效提高标注系统的性能.
In order to improve the performance of automatic image annotation system, a new optimization method based on optimal tag sets is proposed. Firstly, the image is represented by bag-of-word and annotated by CTM model. Then, the correlation between words frequency factor and heuristic iterative algorithm is used to optimize the label set, which can improve the accuracy of label words greatly. Experiments on Corel5K dataset validate that the proposed method can offer better annotation effect than some other annotation methods, such an LDA and CTM model.
出处
《复旦学报（自然科学版）》
CAS
CSCD
北大核心
2014年第1期147-153,162,共8页
Journal of Fudan University：Natural Science
基金
国家自然科学基金项目(50808025)
湖南省科技计划项目(2012FJ3021)
湖南省教育科学"十二五"规划课题(XJK012CGD022)
湖南省普通高等学校教学改革研究资助课题(湘教通【2012】401号544)
