Research And Application Of Single Image De-raining Based On Generative Adversarial Network

Image de-raining is an important task in the field of computer vision,which is concerned by more and more people.Rain can bring visual degradation to captured images and videos.Especially in heavy rain,the rain stripes may cause serious occlusion of the background scene.The accumulation of rain stripes and water particles form a veil on the background,which significantly reduces the contrast of the scene and reduces the visibility.Human vision and many computer vision algorithms are affected by this degradation,because most of these algorithms assume that the weather is clear,without the interference of rain stripes and Rain Water accumulation.Therefore,recovering image background information from rain,called image de-rain,is very necessary in many practical applications.The network with better feature learning and feature expression ability in the deep neural network is to generate the adversarial network,so in this paper,we choose the generated adversarial network as the basic network structure of the single image rain removal model.The idea of confrontation game training is used to generate a clear image of no rain stripes.For this reason,the research of rain removal based on generating a single image of countermeasure network is carried out in this paper.The main work includes:(1)Based on the features of generating adversarial networks and single image rain removal,this paper designs a neural network model for single image rain removal.In the part of the generating network,the coding and decoding structure is selected as the generating structure of the rain removal image.at the same time,in order to reduce the computing resources and computer time of the neural network model as much as possible without reducing the performance,the convolution operation is mainly used in the model.(2)Based on the attention generation,the method of adversarial network de-raining was improved.Through analysis,it was found that the method can remove rain drops in the image very well but remove the rain lines in the image poorly.In affecting the overall quality of the image,raindrops and rain stripes are different.Because most raindrops are transparent,some background information can still be reflected by raindrops,thus providing valuable information for image restoration.Aiming at this point,this paper adds the rain line extraction module to the network structure to improve the attention to the rain line in the image,and to add color loss in the loss function design to make the generated image more real.Through the comparison,the improved method has certain advantages.(3)Using the improved method of attention generation adversarial network rain removal,a single image rain removal system based on We Chat Mini Programs mobile terminal is designed and implemented.The single image rain removal system is analyzed and described from the requirement analysis,system outline design,system detailed design and implementation and system testing.Finally,the test results show the availability and practicability of the system.