An Image Defogging Method Based on Depth CNN Network
Junying Ren, Jin Cheng, Xiaoyuan Wang · 2022
A defogging method to remove the haze from the images of video surveillance system for the high-voltage transmission lines is addressed in this paper. Rain, smog and glare are the leading causes of haze images. To identify the condition of the transmission lines, the video surveillance system need to remove the haze effectively. With CNN network, a haze removal method to defog the image is designed. Firstly, the regression function of CNN network is used to make a difference between positive and negative samples. The foggy image degradation model is chosen for CNN. The foggy images feed into the network, and the network is used for learning and processing. Finally, the CNN network are trained iteratively and learns to transform the fogged negative samples regressively to positive samples in the network. In each iteration, the semantic information of the image between the context and the original image is effectively utilized. Results of experiments show that the network has obtained good processing ability and can be applied to real fogged images of high-voltage transmission lines.