A comprehensive analysis on single image-based deraining using generative adversarial network

Deraining is a process by which we can get a transparent image by removing raindrops from a rainy image. In the rainy time visibility of any scene decreases as vision property is affected by the rain. Recently generative adversarial network (GAN) is getting popular in the visual enhancement of hazy, dusty, and noisy images. It is essential to know the effectiveness of the diverse GAN algorithms in the natural rainy situations of different intensities. From this perspective, the present paper describes a comprehensive study on four single-image state-of-the-art GAN models, such as attentive GAN, cGAN, DHSGAN, and Cycle GAN for deraining. The experiment is done using the standard dataset consisting of real-world rainy images and the results are evaluated both objectively and subjectively. We have found somehow mixed results based on quantitative metrics and comparatively satisfactory results by the cGAN based on visual analysis.