基于自监督学习的光场空间域超分辨成像
Self-Supervised Learning for Spatial-Domain Light-Field Super-Resolution Imaging
梁丹 1张海苗 1邱钧1
作者信息
- 1. 北京信息科技大学理学院,北京 100101
- 折叠
摘要
针对光场成像的空间域图像分辨率限制,提出一种基于自监督学习的空间域图像超分辨成像方法.利用深度学习中的自编码器,对全部光场子孔径图像同步进行空间域的超分辨重构.设计一种基于多尺度特征结构和全变差正则化的混合损失函数,约束模型输出图像与原始低分辨率图像的相似度.数值实验结果表明,所提方法对噪声具有抑制作用,在光场成像的不同数据集上的超分辨结果平均值超过基于有监督学习的光场空间域超分辨方法.
Abstract
This paper proposes a self-supervised learning-based method for the super-resolution imaging of spatial-domain resolution-limited light-field images.Using deep learning self-encoding,a super-resolution reconstruction of the spatial-domain is performed simultaneously for all light field sub-aperture images.A hybrid loss function based on multi-scale feature structure and total variation regularization is designed to constrain the similarity of the model output image to the original low-resolution image.Numerical experiments show that the newly proposed method has a suppressive effect on noise,and the resultant average super-resolutions for different light field imaging datasets exceed those of the supervised learning-based method for light field spatial domain images.
关键词
光场/超分辨/自监督学习/深度学习Key words
light field/super-resolution/self-supervised learning/deep learning引用本文复制引用
基金项目
国家自然科学基金(12101061)
国家自然科学基金(61931003)
出版年
2024
NETL
NSTL
万方数据