基于超像素分割的新闻图像自动语义标注算法.pdfVIP

中国科技论文在线 基于超像素分割的新闻图像自动语义标注 算法# 连洁，蒋晓悦，吴俊，冯晓毅** 5 （西北工业大学电子信息学院，西安 710129 ） 摘要：基于图像场景理解以及语义检索的需要，图像的自动语义标注成为了近几年来非常活 跃的一个研究领域。但由于机器与人类之间存在着“语义鸿沟” ，让机器用人类的思维理解图 像本身存在着一定的难度。同时根据目前的研究成果，室内场景图像的语义标注准确率大约 10 只达到了室外场景的一半，室内图像自动语义标注还有很大的提升空间。因此，本文以新闻 图象作为语义标注的研究对象，新闻图像多为室内场景并且内容较丰富。本文提出综合使用 语义标注中的概率模型与分类模型，并利用马尔科夫随机场为语义标注添加上下文约束的算 法。该算法可有效提高图像语义标注的准确率，并实现室内场景图像的自动语义标注。 关键词：语义标签；新闻图像；图像超像素分割；几何标签；马尔科夫随机场模型 15 中图分类号：TP391 Research on Automatic News Images Annotation Methods based on Superpixel LIAN Jie, JIANG Xiaoyue, WU Jun, FENG Xiaoyi 20 (School of Electronics and Information, Northwestern Polytechnical University, Xian 710129) Abstract: Due to the demand of scene understanding and semantic retrieval of images, automatic image annotation has become a hot research topic in recent years. Due to the-state-of-the-art, the image annotation accuracy of indoor scenes and outdoor scenes varies tremendously. The accuracy of indoor scenes annotation is only half of that of outdoor scenes, so the annotation 25 technique for indoor scene images is far from mature. We take News images as the research object for the reason that most of them belong to indoor scenes and their content is relatively rich. This paper combined the probabilistic model and the classification model to achieve automatic image annotation. And then Markov Random Field is used as contextual constrains to promote the annotation performance. 30 Key words: semantic annotation; news images; the probabilistic model; the classification model; Markov Random Field 0 引言 语义网(semantic wed)是指以机器为中心，接入其中的计算机或其他电子设备拥有很高 35