利用Ontology改进的自动化图像标注方法

Improved method of automated image annotation based ontology

图像检索系统大多是利用图像的底层特征如颜色、纹理和图像来分析图像,没有考虑图像内容及其对象的内容语义,导致对图像的理解不佳。为使系统能更准确的理解图像中的对象及其深层语义,分析了目前图像标注的优缺点,提出了一种以底层特征为基础,利用本体论建构的知识辅助计算机分析图像中实体对象,判断对象与对象间在现实世界中存在的合理相关性,进而对图像进行标注。实验结果显示加入本体论辅助标注图像大大提高了图像识别的准确性。

The past image retrieval systems utilize low-level featrues such as color, texture and shape, to analyze image, but this does not take semantics of content and objects in images into account which usually leads to misunderstanding of images. To understand objects in the images accurately and its deep semantic by computer, the advantages and disadvanta- ges in present image annotation are analyzed, then an improved method is presented which applies semantics of content on analysis of physical objects in images, so computers can accurately detect objects, deduce the relations between objects and extract the underlying semantics. Comparing between conventional low-level annotations and our newly proposed ontolo- gy-based annotations, ontology enhances computer~ s comprehension for images, and the accuracy of object recognition is also increased.