Image Defogging Method Based on Improved CycleGAN Model
The image capture under foggy conditions is not clear enough,which seriously affects the quality of sub-sequent computer vision tasks.Therefore,it is of practical significance to clear the image taken on hazy days.The ex-isting CycleGAN dehazing model fails to make full use of the feature information in the process of encoding and deco-ding,and the quality of the dehazing image obtained is not high enough.In the method,the generator introduces multi-scale information through the designed feature fusion module,the discriminator adopts the network shared double discriminator strategy,and the loss function uses two kinds of adversarial loss combined with the improved cycle con-sistency loss to enhance the model's dehazing performance and image quality.Comparison experiment and ablation ex-periment verify the performance of the improved model.
NETL
NSTL
万方数据
