A Review on Image Dehazing Using Generative Adversarial Network (GAN)
Anuradha K. Shrimawale, Jagdish D. Kene · 2025
The revolution in computer vision techniques and deep learning has played an important role. Among all challenges faced in this field, image dehazing remains a tricky problem. Haze can be caused by adverse or foggy weather conditions, unsuitable lighting, and pollution. The problem increases in intensity, especially in extreme conditions. A deep learning-based approach was developed to solve problems. But learning-based methods still face some problems. They are poor at taking out haze from outdoor images with variable training data due to insufficient use of a variety of training data. This paper presents a series of dehazing techniques, and explores, in particular the GAN-based approach. These GAN based approaches have been introduced to resolve the issues facing image dehazing. GANs are trained using a large collection of haze free and hazy images for the removal of haze from degraded ones and improve computer vision performance in hazy situations. GAN-based image dehazing is widely accepted as the best method for real time dehazing. The present paper qualitatively analyzes different approaches and discusses their strengths and weaknesses. In addition, this review discusses the mainstream benchmarks, existing challenges, performance metrics, and applications in image dehazing research.