Low-Quality Image Enhancement Algorithm for Industrial Scenes Based on Generative Adversarial Networks
叶旭辉 1倪蔚恒 1陈燕 2尹芹凯 1张道德1
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 湖北工业大学机械工程学院,武汉 430068
2. 武汉东湖学院机电工程学院,武汉 430212
折叠
摘要
针对工业场景下图像模糊、分辨率低、边缘细节不明显等问题,提出一种基于生成对抗网络的低质图像增强算法.首先,设计退化网络获得与真实场景更为接近的低质图像,以此与现实高清图像获得特征映射关系;其次,在使用密集残差块(residual in residual dense block,RRDB)的基础上添加卷积注意力模块,增强RRDB网络的特征表达能力,以有效地捕获关键特征信息;最后,设计边缘增强网络模块结合改进的RRDB作为生成器,图像细节信息的捕捉与还原能力得到显著提升,并与判别器对抗生成更高质量的图像.实验结果表明,相较于现有常用的图像增强算法,所提算法能有效提升工业场景图像清晰度、保留图像细节并减少失真.定量指标峰值信噪比平均提升10.45%,结构相似性平均提升15.92%,运行速度快,能满足工业生产需求.
Abstract
Aiming at the problems of image blurring,low resolution and unclear edge details in industrial scenes,this paper proposes a low-quality image enhancement algorithm based on generative adversarial net-works.Firstly,the degraded network is designed to obtain low-quality images closer to the real scene,so as to obtain the feature mapping relationship with the real high-definition images.Secondly,on the basis of u-sing residual in residual dense block(RRDB),a convolution attention module is added to enhance the fea-ture expression ability of RRDB network,so as to effectively capture key feature information.Finally,the edge enhancement network module is designed and combined with the improved RRDB as the generator,the ability of capturing and restoring image detail information is significantly improved,and the higher qual-ity image is generated by confrontation with the discriminator.The experimental results show that compared with the existing commonly used image enhancement algorithms,the proposed algorithm can effectively im-prove the clarity of industrial scene images,preserve image details and reduce distortion.The peak signal-to-noise ratio of the quantitative index is increased by 10.45% on average,and the structural similarity is in-creased by 15.92% on average.The operation speed is fast and can meet the needs of industrial production.
维普
万方数据