Image Dehazing using SR Networks

Project Overview

This project explores advanced techniques for Image Dehazing, focusing on two key directions: reducing computational costs through joint super-resolution training and enhancing accuracy using depth information from stereo and LiDAR systems.

Core Techniques

1. Efficient Dehazing via Super-Resolution

Awarded the Best Paper at IEEE IS3C 2023, this work investigates the synergy between dehazing and Super-Resolution (SR). By employing a Joint Training strategy, we reduce the computational overhead typically required for high-quality restoration, making it suitable for real-time ADAS applications.

2. Depth-Aware Dehazing with LiDAR

Presented at IEEE ICASI 2023, this research leverages Stereo Depth Estimation Networks and LiDAR-Assisted Cameras. The fusion of precise depth data helps the model better estimate haze density (transmission map), leading to cleaner more physically accurate dehazing results.

Related Publications