Image De-Hazing Using Deep Learning Approach

Image de-hazing has always been a challenging task in the domain of Computer vision since the beginning. Sometimes the images that are taken in certain atmospheric conditions turn out to be blurred and hazy, and of low quality because of the adverse environmental conditions, resulting in foggy and hazy capture. This, in turn, creates a lot of trouble to identify or recognize the objects or key items in the captured images. This causes many issues, particularly for the researchers working in the domain of Computer vision, who mainly depend upon the prominence and perceptibility of the captured images as the study data. In this research article, a deep learning algorithm comprising of generative adversarial network architecture has been implemented to de-haze the image. The actual de-hazing methods and processes involve the use of per-pixel loss, which during the analysis period creates a big problem even if there is a difference of only one pixel in the whole of the image or even if the captured images are seemed to be perceptually alike. The function of the perceptual gains the high-level characteristics of the captured images by making use of the pre-trained algorithms on ImageNet, which minimizes or diminishes the challenges that occur in the perpixel loss function