Fog Restoration in Hazy Images using Deep Transfer Learning
Abstract
Issues like low vision and the distortion of spectral and spatial information in photographs can be brought on by unfavorable atmospheric circumstances including smog, rain, overcast sky, fog, and smoke. This piece looks at images taken in a hazy environment. While de-smogging requires veil and transmission map data, optical imaging systems are limited to capturing smoggy images. To restore blurry photographs, effective optical information prediction is necessary. Research has shown that optical information anticipated by the dark channel prior technique frequently produces subpar images in photographs with areas of higher brightness, noticeable pollution gradients, or textured information. In order to estimate optical information, this study makes use of Deep Transfer Learning (DTL) and Optimized Gradient Prior Prediction (OGPP).We gathered benchmark images of both smoggy and smog-free environments in order to train the DTL model. The OGPP technique uses the smog gradient, which the DTL model is taught to anticipate, to recover pictures free of pollution. A comparison analysis shows that when it comes to recovering hazy photos, the OGPP restoration model performs better than other approaches.