Main content area

Superpixel segmentation for hyperspectral images based on false colour composition with colour histogram driving

Lin, Lianlei, Yang, Jingli, Zhou, Zhuxu
International journal of remote sensing 2019 v.40 no.1 pp. 307-331
algorithms, color, data collection, hills, hyperspectral imagery, remote sensing, system optimization
Hyperspectral images (HSIs) segmentation has gradually become an important basis for HSIs processing, such as the classification and unmixing. In this paper, we proposed a superpixel segmentation algorithm based on principal component (PC) weighted false colour composition (FCC) with colour histogram driving (FCC-CHD). First, the dimensionality of HSIs is reduced by using FCC algorithm based on weighted PC, so that the main spectral information of HSIs is transformed to colour information of false colour images. To scale difference of ground objects in HSIs, we propose colour histogram driving (CHD) function to guarantee the accuracy of metric functions and computational efficiency. In segmentation process, the hill climbing optimization (HCO) algorithm is used to transform a global optimization problem into a local optimization problem, ensuring the high efficiency of the algorithm. Finally, parameters selection experiments provide reasonable parameter selection references, and the comparison experiments on two datasets demonstrate that the proposed algorithm is more effective and accurate than other superpixel segmentation methods.