Video image defogging recognition based on recurrent neural network

Xiaoping Jiang, Jing Sun, Chenghua Li, Hao Ding

Abstract: Fog and haze make the image degraded, and seriously affect the normal operation of the information system in the fields of military, transportation, and safety monitoring, under this condition, the image defogging is of great significance. In order to meet the needs of real-time processing of existing video's defogging process, a recognition algorithm based on recurrent neural network is proposed in this paper. At present, the mainstream image defogging algorithm mainly uses a variety of fog related color features, however, different color prior knowledge often has its own scene limitation. We use sparse automatic coding machine to extract the texture features of the image, and extract all kinds of fog related color features. Then, we use the recurrent neural network to implement sample training process, and we obtain the mapping relationship between texture structure features and color features and scene depth, and then we estimate the scene deep map of fog images. Finally, the atmospheric scattering model is used to recover the fog-free image according to the scene deep map. Experiments show that the proposed algorithm can effectively obtain the scene depth of the image, and recover the ideal fog-free image. Robustness has been tested and guaranteed through the numerical simulation.