An Unsupervised Image Dehazing Network Based on Dual Discriminators and Contrastive Reconstruction

Abstract

To tackle the challenges of uneven haze distribution, blurring, and color distortion in complex scenes, this paper proposes an unsupervised image dehazing algorithm based on dual discriminators and bidirectional contrastive reconstruction. The framework integrates an end-to-end dehazing network with a parameter estimation module, featuring sub-networks for atmospheric light and transmission map estimation. Leveraging parallel attention, multi-scale learning, and a variational autoencoder, the model effectively captures key features and adaptively estimates haze parameters, enhancing clarity and realism. Experiments on benchmark datasets demonstrate that the proposed method outperforms existing unsupervised approaches in both visual quality and quantitative metrics, while also showcasing strong robustness and practical applicability.