A New Hyperspectral Multi-Level Synthetic Hazy Image Dataset for Benchmark of Dehazing Methods
Bilge Yazıcı, Yücel Çimtay, Bedrettin Çetinkaya · 2023
In this study, a new hyperspectral-multi-level hazy image dataset is presented. There are many single-level color and several multi-level color hazy image datasets in the literature. However, there is a lack of hyperspectral multi-level hazy image dataset. SHIA dataset is the only hyperspectral multi-level hazy image dataset in the literature. The main goal of this study is to present the new hyperspectral multi-level synthetic hazy image dataset to contribute to the related dehazing literature. This dataset is created by using 5 different scenes. The hyperspectral images with 10 nm wavelength bandwidth were collected from an existing dataset: Real-World Hyperspectral Images Database. For each image, the state-of-the-art depth estimation method: "Dense-Depth-Master" is used and depth maps were obtained. By changing the haze level parameter, "Haze-synthesize" is used to add haze to each single band image of the hyperspectral image data. In this study, for benchmark of different state-of-the-art dehazing methods, we conducted tests on the hyperspectral hazy images.