Fast Hyperspectral Image Dehazing With Dark-Object Subtraction Model

Abstract

Haze in hyperspectral images is a common phenomenon, which severely degrades the quality of the acquired data and limits its applications. In this letter, a dark-object subtraction model is proposed for haze removal of hyperspectral images, which mainly is composed of three steps. First, a haze density map is estimated according to the haze characteristic in different spectral channels. Then, we design a saliency measure method to automatically calculate haze abundance of different channels. Finally, the haze-free image is obtained through solving the dark-object subtraction model. Experiments on real and simulated datasets demonstrate that our method consistently outperforms other state-of-the-art dehazing techniques in terms of reconstructed performance and computational cost.