Influence of Spectral Metrics on the Graph-Based Segmentation of Hyperspectral Images

摘要

Graph-based representation, which has recently been used for hyperspectral image segmentation, effectively allows the simultaneous exploitation of the spatial and the spectral data of these images. This representation may, however, depend on several aspects which affect the segmentation process.In this paper, we study the impact of the spectral metric used to construct a weighted graph in hyperspectral image segmentation. We employ the Normalized graph cut and the multiphase level set on graphs as segmentation algorithms. The experimental results demonstrate the important influence of spectral distance functions on the quality of hyperspectral image segmentation results.