摘要
针对传统数字图像去雾算法容易受到先验知识制约、参数估计困难等问题,提出了一种基于改进AOD-Net的端到端图像去雾算法。从网络结构和损失函数两部分对AOD-Net算法进行优化;采用高斯图像金字塔模型设计了包含3个尺度图像的网络结构,提升去雾图像的边缘细节质量;综合视觉感受设计了包含SSIM损失函数和均方差损失函数的混合损失函数,提升去雾图像的亮度和对比度。实验结果表明,改进AOD-Net算法在图像去雾的主观视觉效果和客观数值结果上,相比较其他经典去雾算法均有着更好表现。
An end-to-end image defogging algorithm based on improved AOD-Net is proposed.The traditional digital image defogging algorithm is easily restricted by prior knowledge and parameters are hardly estimated,in order to solve this problem,the AOD-Net algorithm is optimized from both the network structure and loss function.The Gaussian image pyramid model is used,and the network structure containing three scale images is designed to improve the edge detail quality of the defogging image.A mixed loss function including SSIM loss function and mean square error loss function is designed to promote the brightness and contrast of the defogging image.Compared with other traditional dehazing algorithms,the experimental results show that the improved AOD-Net algorithm has excellent performance in the subjective visual effect and objective numerical results of image dehazing.
作者
马学条
MA Xuetiao(School of Electronics and Information Engineering,Hangzhou Dianzi University,Hangzhou 310018,China)
出处
《实验室研究与探索》
CAS
北大核心
2023年第7期38-43,共6页
Research and Exploration In Laboratory
基金
杭州电子科技大学高教改革重点项目(ZDJG202101)。
关键词
图像去雾
网络结构
损失函数
image defogging
network structure
loss functions
