多尺度生成对抗网络单幅图像去雨方法研究

Research on Rain Removal Method for Single Image of Multi-scale Generative Adversarial Network

随着人工智能的迅猛发展,基于图像识别、物体检测、目标跟踪等技术的计算机视觉系统被广泛应用,然而受到特殊天气状况(如雨天)的影响,该类系统采集到的图像质量受损,直接导致其性能下降,给相关领域带来严重损失。因此,对图像去雨的研究受到众多学者的关注。近几年,深度学习在计算机视觉领域大放异彩,基于生成对抗网络提出了一个端到端的单幅图像去雨网络,引入多尺度采样和残差网络的思想,实验证明该网络能较好地实现图像去雨,提高了图像去雨的性能。

With the rapid development of artificial intelligence,computer vision systems based on image recognition,object detection,target tracking and other technologies are widely used.However,due to the impact of special weather conditions(such as rain),the quality of the images collected by this type of system is impaired,which directly leads to reduce its performance and brings serious losses to related fields.Therefore,the research on removing rain from images has attracted the attention of many scholars.In recent years,deep learning has shined in the field of computer vision.This paper proposes an end-to-end single-image rain removal network based on a generative adversarial network,and introduces the idea of multi-scale sampling and residual network.Experiments prove that the network can achieve better image removal from rain and improve the performance of image removal from rain.