华中科技大学学报（自然科学版）2025，Vol.53Issue(3)：1-7,7.DOI:10.13245/j.hust.250671
一种基于对极约束注意力的光场图像超分辨率方法
An epipolar constraint attention method for light field super resolution
摘要
Abstract
To solve the spatial resolution limitations of light field imaging,caused by spatial-angular multiplexing and lead to spatial resolution degradation,a deep learning-based light field spatial super-resolution method was proposed,utilizing an epipolar constraint attention mechanism.By analyzing the disparity distribution characteristics of the light field and the combination of the principle of attention networks,an epipolar constraint attention module was designed to confine attention within a specific range of the light field's epipolar-plane image,thereby enhancing light field super-resolution relevant features and suppressing irrelevant ones.Experimental results demonstrate that this range can achieve the best performance when matched with the disparity range of the input light field,with a certain degree of interpretability.The experimental results also show that the proposed method outperforms 10 comparative methods in terms of average peak signal-to-noise ratio and structural similarity and exhibits superior visual quality.关键词
光场图像/图像超分辨率/对极约束/深度学习/注意力机制Key words
light field image/image super resolution/epipolar constraint/deep learning/attention mechanism分类
信息技术与安全科学引用本文复制引用
杨敬钰,金满昌,胡昆澍,刘高生..一种基于对极约束注意力的光场图像超分辨率方法[J].华中科技大学学报（自然科学版）,2025,53(3):1-7,7.基金项目
国家自然科学基金资助项目(62231018). （62231018）
