Improving dehazing results for different weather conditions using Guided Multi-Model Adaptive Network (GMAN) and Cross-Entropy Deep Learning Neural Network (CE-DLNN)

Abstract

In computer vision, image dehazing remains a pivotal challenge, especially in accommodating diverse weather conditions that greatly impact visibility and image quality. The development of deep learning algorithms for image dehazing has been a prominent area of research in recent years. Two such methods are the Cross-Entropy Deep Learning Neural Network (CE-DLNN) and the Guided Multi-Model Adaptive Network (GMAN), which have shown promising results in removing haze from images. In this paper, the performance of these two methods is compared in terms of PSNR, SSIM, and MAE on dehazed images. It is found that GMAN outperforms CE-DLNN in these metrics, producing dehazed images with higher PSNR, SSIM, and lower MAE. Additionally, an investigation into the combination of GMAN and CE-DLNN demonstrates further improvements in the performance of both methods, resulting in higherquality dehazed images with enhanced details and textures. These findings substantiate the potential of utilizing deep learning-based approaches for image dehazing and highlight the benefits of synergistically integrating various methodologies to optimize performance.