Nighttime Road Scene Image Enhancement Based On Cycle-consistent Generative Adversarial Network

This article discusses various techniques for enhancing nighttime road scene images, including histogram equalization, Retinex theory, deep learning, and hybrid methods. The authors also propose an improved loss function that combines least squares loss and unsupervised perceptual loss to evaluate generative and discriminative networks. This approach effectively improves image quality by aligning sample features with the decision boundary.