EPQ-GAN: Evolutionary Perceptual Quality Assessment Generative Adversarial Network for Image Dehazing

Single image dehazing is a challenging issue with the goal to improve the quality of hazy images for computer vision applications and surveillance systems. The majority of current techniques aim to restore clear images from hazy images by approximating the transmission map and global atmospheric light. However, inaccurate estimation of these factors results in unrealistic outcomes. To overcome these challenges and to produce realistic images, we implemented a new approach called the Evolutionary Perceptual Quality Assessment Generative Adversarial Network (EPQ-GAN). The EPQ-GAN comprises a novel Evolutionary Generator and Discriminator. The Evolutionary training of the Generator can improve generative performance; hence, to train the Evolutionary Generator, the proposed method adopted the Differential evolution (DE) algorithm and the entire GAN network is trained with Perceptual loss from the discriminator, quality assessment loss (PSNR Loss) and adversarial loss. The proposed EPQ-GAN has superior performance compared to other state-of-the-art approaches, as evidenced by both qualitative and quantitative examination of several benchmark datasets.