基于AOD-Net改进的轻量级图像去雾算法  

Improved lightweight image defogging algorithm based on AOD-net

作　　者：鄢江涛 YAN Jangtao(School of Computer and Information Science,China Three Gorges University,Yichang 443000,China)

机构地区：[1]三峡大学计算机与信息学院,湖北宜昌443000

出　　处：《长江信息通信》2023年第7期40-43,共4页Changjiang Information & Communications

摘　　要：针对雾天图像模糊的问题,提出一种改进的AOD-net网络。为了解决AOD-Net损失函数对图像亮度和对比度关注度不足,文章对AOD-Net进行了优化,并改进了损失函数为SSIM-L1。处理合成雾数据(NYU2),与DCP、Dehaze-Net和AOD-Net相比,文章方法的平均PSNR为23.865。本文方法的HVS-MaxPol_2指标在测试处理真实有雾图像领先。实验结果表明,本文提出的改进AOD-Net网络方法提升了去雾后的视觉效果。An improved AOD-net network is proposed to solve the problem of blurred images in foggy days.In order to solve the problem that AOD-Net loss function does not pay enough attention to image brightness and contrast,this paper optimizes AOD-Net and improves the loss function to SSIM-L1.When processing synthetic fog data(NYU2),the average PSNR of the proposed method is 23.865 compared with DCP,Dehadce-Net and AOD-Net.The HVS-MaxPol_2 index of the proposed method is leading in the test processing of real fog images.Experimental results show that the improved AOD-Net network method proposed in this paper improves the visual effect after fog removal.

关 键 词：图像去雾 深度学习 HVS-MaxPol AOD-Net 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]