The Impact of Haze Non-Homogeneity on the Recent Image Dehazing Methods

Abstract

Haze is a common atmospheric phenomenon that affects the visual quality of the recorded images. Since in general the existing dehazing techniques assume a homogeneous haze distribution, this work analysis the impact of haze non-homogeneity on several recent image dehazing techniques. We perform a comprehensive evaluation using our recent image dataset introduced this year at CVPR NTIRE 2021 workshop. HH-HAZE21 dataset contains 35 non-homogeneous hazy images and their corresponding haze-free images captured from the same scene. The hazy images have been generated using a special setup that includes a professional haze generator. The qualitative and quantitative comparative results demonstrate that images with non-homogeneous haze introduce new challenges for the existing image dehazing techniques.