摘要
针对光线通过介质会产生散射现象,提出了一种基于神经网络的端对端图像去散射方案,对于散射退化的图片进行退散射的处理,此方案不需要复杂的光学设备,应用场景广泛,在仿真和实际实验中都取得了预期的结果。提出了一套利用拍摄屏幕来建立散射退化图片和无散射图片的数据集的方法,且这一方法在许多其他的图像处理图像恢复工作中有应用价值。
The image quality will decay when light goes through the scattering media or reflect on the rough surface.In this paper,we purpose an end-to-end deep learning descattering method.This method does not need complex optical device and perform very well in complex scene.This method has a state-of-art performance in both the simulation dataset and real experiment dataset.In this paper,we also creatively put forward a pipeline to collect the hazing and ground truth dataset by capturing the screen,which can also be used in many other field.
作者
漆建军
Qi Jianjun(School of Information Science and Technology,Guangdong University Forien Studies,Guangzhou 510006,China)
出处
《信息技术与网络安全》
2020年第9期49-55,共7页
Information Technology and Network Security
关键词
去散射
深度学习
端对端
图像恢复
descattering
deep learning
end to end
image reconstruction
