Comparison of Superpixel Algorithms on Hyperspectral Images

Abstract

The hyperspectral images are conventionally classified using spectral information. Spectral information does not include the neighborhood relationship of the pixels. In this paper we classfy the hyperspectral images by using state-of-the-art superpixel methods and compare their performances. Since the superpixels take into account neighborhood relationship of the pixels, we benefit from the cooperation of spatial and spectral information. The superpixel-based classification method outperforms pixel-wise methods.