基于深度学习的光场超分辨率算法综述  

Review of Light Field Super-Resolution Algorithm Based on Deep Learning

作　　者：熊娅维 王安志 张凯丽 Xiong Yawei;Wang Anzhi;Zhang Kaili(School of Big Data and Computer Science,Guizhou Normal University,Guiyang 550025,Guizhou,China)

机构地区：[1]贵州师范大学大数据与计算机科学学院,贵州贵阳550025

出　　处：《激光与光电子学进展》2024年第18期28-39,共12页Laser & Optoelectronics Progress

基　　金：国家自然科学基金地区科学基金项目(62162013);贵州师范大学学术新苗基金项目(黔师新苗[2022]30号)。

摘　　要：光场图像分辨率低的原因之一是光场空间分辨率和角度分辨率之间存在相互制约。光场超分辨率技术旨在从低分辨率光场图像中重建出高分辨率光场图像。基于深度学习的光场超分辨率方法通过学习高、低分辨率光场图像之间的映射关系来提升图像的质量,突破了传统方法计算成本高、操作复杂的限制。本文对近年来基于深度学习的光场超分辨率技术研究进展进行了全面综述,梳理了网络框架和典型算法,并进行了实验对比分析。最后,总结了光场超分辨率领域面临的挑战,并展望了未来可能的发展方向。The trade-off between spatial and angular resolutions is one of the reasons for low-resolution light field images.Light field super-resolution techniques aim to reconstruct high-resolution light field images from low-resolution light field images.Deep learning-based light field super-resolution methods improve the quality of images by learning the mapping relationship between high-and low-resolution light field images.This advantage breaks through the limitations of traditional methods with high computational cost and complex operation.This paper provides a comprehensive overview of the research progress of deep learning-based light field super-resolution technology in recent years.The network framework and typical algorithms are examined,and experimental comparative analysis is conducted.Furthermore,the challenges faced in the area of light field super-resolution are summarized,and the future development direction is anticipated.

关 键 词：光场 图像超分辨 图像修复 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]