NTIRE 2024 Dense and Non-Homogeneous Dehazing Challenge Report

This study examines the results of the NTIRE 2024 Challenge on Dense and Non-Homogeneous Dehazing. Innovative methods were introduced and tested using a new image dataset named DNH-HAZE. The DNH-HAZE dataset comprises 50 pairs of authentic outdoor images showcasing dense and non-homogeneous haze alongside corresponding haze-free images of identical scenes. The haze was simulated using a professional setup designed to mirror real-world hazy conditions. The competition attracted 374 participants, with 16 teams presenting solutions for the final evaluation phase. The proposed solutions showed the leading edge of image dehazing technology.

1. Introduction

Haze is a naturally occurring phenomenon that can greatly reduce visibility in a scene with increasing distance, leading to reduced image quality. This atmospheric effect is triggered by the presence of small airborne particles that affect environmental properties. As a result, hazy scenes often exhibit low contrast, reduced saturation, altered colors, and increased noise.