Improved Resolution of Dehazed Images with Dark Channel Prior and Super Resolution GAN

Extracting information from an image becomes difficult when the scene is disrupted due to bad atmospheric conditions. Suspended particles like fog, rain and dust creates haze in the scene. In many real time situations vision algorithms work on clean and sharp images. This makes the task of vision algorithms difficult to operate and get correct results. Moreover, after dehazing an image still a lot of information gets missing because recovery is not perfect. To fill the information gaps resolution of the image needs to be improved. Haze removal and image resolution improvement is a difficult task and this problem is poorly posed. Over the years many methods have come to address the problem of haze and also improvement of resolution of image. However, they have been addressed independently and in many real time environments resolution of dehazed image is low due to which many vision algorithms fail. In this paper, we propose a method combining Dark Channel Prior and Super Resolution GAN removing haze and improving resolution of image simultaneously followed by working principle with proper discussion of algorithm accompanied with visual representation of experimental results.