可调型Ising模型及其在高分辨率影像主干道提取中的应用

An adjustable-Ising model and its application in the major road extraction from high-resolution imagery

结合基于随机场理论的影像分类技术,提出高分辨率影像的主干道提取算法.在影像分类时,通过Ising模型将像元空间上下文关系融入到分类中.为避免由Ising模型的"边缘模糊"现象,对该模型进行了改进,并提出可调型Ising模型.新模型通过对边缘处和区域内的像元施以不同强度的类型平滑操作,不仅避免了"边缘模糊"现象,而且还有效地屏蔽了主干道区域内噪声和细节信息的干扰,提高了分类结果中该区域的完整性.最后,根据几何形状特征在分类结果中对主干道进行提取.结果表明:相比基于K均值聚类法和基于Ising模型分割算法的提取结果,本文算法在区域完整性和边缘类型判别精确性等方面均有较大程度的提高.

This paper proposes an algorithm on the main road extraction from the high-resolution imagery by using the image classification technology based on the random field theory.During the image classification,the context of pixels is incorporated by using the Ising model.In order to avoid the ＂blurred edge＂ phenomenon produced by the Ising model,an adjustable-Ising model is proposed.The new model can smooth the pixels on the edges or inside the major road,which represent different land cover classes,in different the strength. This not only avoids the ＂blurred edge＂ phenomenon, but also effectively depressed the un -wanted details and noise within the major road, which greatly improves the road integrity. Finally, according to the geometric morphology, the major road is extracted from the result of classification. The results show that the proposed algorithm is more capable in the extraction accuracy than K - means clustering algorithm and the Ising model, as the new model can produce more clear edge with high road integrity.