An efficient defogging network for RAW image sequences with high viewpoint

The image information captured by the camera is seriously lost due to the thick concentration of fog in high altitude. An efficient defogging network for RAW image sequences with high viewpoint, termed LWRAW-ClassifyCycle, is proposed. The network uses RAW data as input to provide higher quality and richer image information. An improved classification module is introduced to reduce the impact of redundant information on the follow-up network and improve the accuracy. The lightweight style migration module is used to solve the problems of image distortion and hue artifact. In addition, a dataset with 1209 high-viewpoint foggy RAW image sequences is captured, including the corresponding JPEG format, to enrich RAW and JPEG data. Extensive experimental results show that this method has better visual effect and real-time performance than the previous methods, and can recover more realistic image details. The dataset will be made public.