GAN-based Multimodal fusion for image dehazing

Haze disturbances degrade image clarity, adversely impacting the performance of vision-based systems in smart cities. To address the issues of blur and colour artefacts in conventional image dehazing methods, this paper proposes a Compensation Generative Adversarial Multimodal Dehazing Network. The Multimodal framework is built on a generative adversarial network architecture and incorporates a compensation modality to mitigate information loss during the dehazing process. The generator consists of two components: a dehazing modaland a compensation modal,which help prevent grid artefacts and enhances feature representation. Experimental results demonstrate that our networrk effectively reduces blur and colour distortion in dehazed images.