基于亚像素和梯度引导的光场图像超分辨率
Light field images super-resolution based on sub-pixel and gradient guide
韦玮 1陈芬 1张华波 1罗英国 1张鹏 1彭宗举1
作者信息
- 1. 重庆理工大学电气与电子工程学院,重庆 400054
- 折叠
摘要
针对光场相机捕获到的光场图像空间分辨率较低等问题,提出一种基于亚像素和梯度引导的光场图像超分辨率方法.设计了多重亚像素信息提取模块,该模块将子孔径图像分为水平、垂直、对角和反对角 4个图像堆,并分别提取各图像堆的亚像素信息.同时,考虑到梯度先验可为预测高频细节提供有效线索,在重建过程中融合了子孔径图像的梯度多重亚像素信息.在 5个公开数据库上的实验结果表明,本文方法不仅在客观指标上普遍优于现有方法,在主观视觉效果上也有更好的表现,边缘纹理细节更加清晰.
Abstract
For the problem of low spatial resolution of light field images captured by light field cameras,a super-resolution method for light field images based on sub-pixel and gradient guide was proposed.A multiple sub-pixel information extraction module was designed,which divided the sub-aperture images into four image stacks:horizontal,vertical,diagonal and anti-diagonal,and extracted the sub-pixel information of each image stack separately.Meanwhile,considering that the gradient prior could provide effective clues for predicting high-frequency details,the gradient multiple sub-pixel information of the sub-aperture images was fused in the reconstruction process.Experimental results on five publicly available databases show that the proposed method not only generally outperforms the existing methods in terms of objective indexes,but also has better performance in the subjective visual effect,which the edge texture details are clearer.
关键词
超分辨率/光场图像/亚像素信息/深度学习Key words
super-resolution/light field images/sub-pixel information/deep learning引用本文复制引用
基金项目
重庆市自然科学基金(cstc2021jcyjmsxmX0411)
重庆市自然科学基金(CSTB2022NSCQ-MSX0873)
重庆理工大学研究生教育高质量发展行动计划(gzlcx20233138)
重庆理工大学研究生教育高质量发展行动计划(gzlcx20233081)
出版年
2024
NETL
NSTL
万方数据