Fast Hierarchical Segmentation of High-Resolution Remote Sensing Image with Adaptive Edge Penalty

作者：Zhang Xueliang^*; Xiao Pengfeng; Feng Xuezhi

来源：Photogrammetric Engineering and Remote Sensing, 2014, 80(1): 71-80.

DOI：10.14358/pers.80.1.71

A fast hierarchical segmentation method (FHS) for high-resolution remote sensing (HR) image is proposed in the paper. FHS is completely unsupervised. It is characterized by two aspects. First, the hierarchical segmentation process is accelerated by the improved linear nearest neighbor graph (LNNG) model and the segment tree model. It runs faster than other existing hierarchical segmentation methods, and can produce multi-resolution segmentations in time linear to the image size. Second, an adaptive edge penalty function is introduced to formulate the merging criterion, serving as a semantic factor. A set of QuickBird, World View, and aerial images is used to test the proposed method. The experiments show that the multi-resolution segmentations produced by FHS can represent objects at different scales very well. Moreover, the adaptive edge penalty function helps to remove meaningless weak edges within objects, enclosing the relation between segments and real-world objects.

• 出版日期2014-1
• 单位南京大学