Image Defogging Algrithom Based On Gene Rative Adversarial Nets

With the progress of science and technology,a large number of outdoor digital images are collected and analyzed for various scientific research and practical production,such as target detection,terrain classification,outdoor photography,etc.However,due to the moisture or the presence of suspended particles in the air,the outdoor image is often accompanied by fog or haze,this caused a lower contrast,loss of part of the scene,the color deviation,and a series of degradation phenomenon,having different degree of effects on scientific research and practical production.Therefore,finding an effective method of digital image to fog is essential and indispensable.In this paper,use CGAN structure,considering the defogging process of a natural fog map as a fog-to-fog style conversion problem of the same scene graph,in combination with the features of the haze image restoration,add tone constraints.The new method solves the following problems compared to the existing methods:(1)In view of the characteristics of the artificial designed model of the current defogging method,a new image defogging method is proposed based on the generative adversarial nets.The method used the generator of generative adversarial nets to generate a defogged image,the experiment proves that this method can achieve satisfactory effect on image defogging.(2)Considering the traditional structure of neural network used to solve the problem of image defogging,need to get a lot of matches paired foggy and clear images of the same scene as the training data set,but it is really a difficult task in practice.This paper designed a generative adversarial nets with special structure,changing the training data set into a large number of unpaired images,greatly reduces the difficulty of obtaining the training data sets.(3)In view of the problem that current defogging algorithm easily causes color distortion,this paper designed a loss function with color constraints to increase the limit to color difference between the foggy image and the generated image.The experiments proved that the new loss function effectively solved the color distortion problem.