基于语义相似度的海量遥感图像可视化研究

针对海量遥感图像如何有效的传达分类结果以实现有效的可视化问题,本文按照分类的语义标注结果的相似度并运用现有的信息可视化技术来实现图像的可视化。首先采用了贝叶斯网络学习的方法进行图像的自动分类标注,然后利用基于图像布局的多维标度算法(Multi-dimensional Scale)以及无需降维的Value and Relation(VaR)技术实现可视化。实验表明本文的方法能够填补图像低层视觉特征和高层语义之间鸿沟,对大量的图像在一个视图内进行有效的浏览,而不造成图像的混乱,并能实现高层次的图像分析。实验的可视化结果是十分有效的。

Research on massive remote sensing image visualization based on semantic similarity

According to the similarity of the semantic annotation, a visualization method of massive remote sensing images is achieved by applying existed information visualization techniques. Image automatic classification and annotation was done based on Bayesian network. Then an image layout based multi-dimensional scaling (MDS) method, and the Value and Relation (VaR) technology that allows effective high dimensional visualization without dimension reduction are used for visualization. Experiments demonstrate the effectiveness and efficiency of proposed approach which can bridge gap between images' low level vision features and high level semantics, perform a large amount of image visualization without clutter, and support users for high level analysis.