摘要
基于关键字的自动图片标注方法,可以更为有效地实现海量图片的管理和检索。然而由于"语义鸿沟"问题,传统的自动图片标注效果往往并不理想。因此,对不精确的标注结果进行优化就显得尤为重要。文中提出一种新颖的图片标注方法。首先,利用基于相关性模型的递进算法得到图片的初始标注结果。然后,利用一种半监督的学习模型,也即随机游动与重新启动算法对得到的初始标注结果进行优化,并选择一定数量的顶端标注作为图片最终的标注。通过在通用Corel图片数据库的实验表明,文中提出的方案可以有效地提高图片自动标注的性能。
Automatic image annotation via keywords is a effective way to manage and retrieve images.However,system performances of existing state-of-the-art methods are often not so satisfactory for the problem of semantic gap.In this paper,a novel approach is proposed.The candidate annotations for a query image are first obtained via a relevance-based progressive algorithm.Then,a semi-supervised learning model,fast random walk with restart algorithm,is utilized to re-rank the candidate annotations and the top ones are reserved as the final annotations.Experiments conducted on the typical Corel dataset shows that the proposed scheme can effectively improve the automatic annotation performance.
出处
《南京邮电大学学报（自然科学版）》
2010年第6期85-88,95,共5页
Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition
基金
南京邮电大学引进人才科研启动基金(NY209018
NY209020
NY207090)资助项目
