LAUSR

Images captured in poor atmospheric circumstances, like smog, rainy, cloudy, fog, smoke, etc., suffer from number of problems such as poor visibility, distortion of spectral ans spatial information, etc. In this paper, we have considered images taken in smoggy environment. However, optical imaging systems provide only smoggy images, but desmogging process require atmospheric veil and transmission map information. Thus, to restore smoggy images, it is required to predict optical information of smoggy images in an efficient manner. From the extensive review, it has been found that the optical information predicted using various channel prior such as dark channel prior may provide poor results especially when images contain brighter regions, large smog gradient, textured information, etc. Therefore, in this paper, a deep transfer learning (DTL) and oblique gradient profile prior (OGPP) is utilized to approximate the optical information. To train DTL, we have obtained benchmark somggy and smog-free images. Thereafter, DTL model is trained. Smog gradient predicted using DTL is used by OGPP model to recover smog-free images. Comparative analysis prove that the proposed DTL-OGPP based restoration model performs significantly better than the competitive restoration models.