摘要
图像语义自动标注成为基于内容的图像检索研究的热点,提出一种改进的SML两级图像语义自动标注方法.首先采用监督多类标注方法 SML对图像进行粗略标注,然后用基于本体的最优语义标注方法(Oostia)对粗略标注的结果进行精细标注,Oostia方法通过4种不同方式对粗略标注关键字进行扩展,充分挖掘图像中丰富的语义信息.文中提出的方法与其它相关方法进行了比较,实验结果表明,该方法优于其它方法.
Semantic auto-annotation of Image becomes research focus in image retrieval Based on content. The two step semantic auto- annotation of image method is proposed. Firstly, a supervised multi-class labeling method (SML) is adopted to coarse annotation for image, then a optimal semantic tag annotation based on Ontology method (Oostia) is employed to fine annotate. There are four ways to extend coarse annotation result in the Oostia method, which can fully mine ample semantic information in images. The proposed method is compared to others, Experiments result show that the proposed method outperforms others.
出处
《小型微型计算机系统》
CSCD
北大核心
2012年第9期2109-2112,共4页
Journal of Chinese Computer Systems
基金
国家自然科学基金重大项目(79816101)资助
关键词
SML
本体
语义标注
图像检索
SML
ontology
semantic annotation
image retri