基于相似性迁移学习的图像标注  

AN IMAGE ANNOTATION METHOD BASED ON SIMILARITY TRANSFER LEARNING

作　　者：周家琪[1] 刘丽[1] 崔晓萍[1] 李方方[1] 

机构地区：[1]山东师范大学信息科学与工程学院,济南250014

出　　处：《山东师范大学学报（自然科学版）》2016年第2期22-26,共5页Journal of Shandong Normal University(Natural Science)

基　　金：国家自然科学基金资助项目(61170145;61373081);教育部博士点基金资助项目(20113704110001)

摘　　要：提出一种基于相似性迁移学习的图像标注算法.首先建立图像间的特征相似度量,引入相似性迁移学习算法,将图像的底层特征相似度量迁移到图像所对应标注词的相似度量,通过统计方法实现图像的自动标注.实验表明,本文方法能够有效提高图像的标注质量,减少噪声干扰.In this paper, we propose an image annotation scheme based on similarity transfer learning. First of all, we establish similarity measures between the features of images. Then we introduce the thought of similarity transfer learning which could transfer the similarity of the image characteristics to the similarity of the image annotation. After that, our work is to realize the image annotation by using the statistical methods. The effectioeness of our image annotation method is proved by the experimental results. This kind of method could improve the quality of image annotation, and reduce the influence of the noise of images.

关 键 词：相似度量 迁移学习 统计 图像标注 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]