Semi-supervised Image Deraining Method Based On Generative Adversarial Network And Convolutional Neural Network

The concept of image rain removal technology is to take photos taken when it rains,eliminate raindrops in the picture,and get the restored photos,which belong to the category of image processing in the field of computer vision,in addition to picture fog removal,super resolution,etc.The essence of image rain removal is to separate the background of the image from the superimposed raindrops,and remove the rain.Restoring the clear image as much as possible is also a major research direction in the field of computer vision,especially in the work of object detection and recognition.This thesis proposes two methods of image rain removal based on deep learning,as follows:(1)An image deraining method based on the combination of GAN-CNN network is proposed,which uses random noise to synthesize complex rain maps,eliminates structural rain streaks by generating the generated network and the discriminant network in the generated adduction network GAN to contend with each other,and uses the convolutional neural network CNN to restore the details in the original image.The simulation results show that this method can better preserve the details of the image while removing the rain lines.(2)Semi-supervised image deraining method based on the combination of GAN-CNN network is proposed.On the basis of semi-supervised matching network,statistical information between the real rain map and the synthetic rain map is used to minimize the difference in feature distribution between the two,so that the rain line distribution of the synthetic rain map is closer to the real rain map.The simulation results show that the proposed method enhances the rain removal performance of the training network on the real rain image.The two methods remove rain from the perspective of supervised and semi-supervised learning,respectively.The first method pays attention to the image removing rain while preserving the texture details of the image,and the second method adds the true rainy image on this basis,saving the training cost and improving the generalization ability of the model.