摘要
近年来,多层网络在图像去模糊领域取得了较大进展,但其性能受限于特征提取和残差连接。为解决这些问题,提出了一种多尺度融合网络(Multi‑Scale Feature Extraction and Fusion Net‑work,MSFN)用于图像去模糊,通过多尺度输入与输出,增强了对图像特征的提取能力。MSFN利用其特征自适应细节增强(Adaptive Detail Enhancement,ADE)模块和跨尺度特征融合(Cross‑Scale Feature Fusion,CSFF)模块,在不同网络深度上捕获不同尺度的特征,优化了特征提取过程,并有效融合了多尺度信息。实验结果表明,所提出的算法在定量分析上表现出色,并且在主观视觉效果上也得到了显著提升,这些结果充分证明了所提网络的卓越性能。
Significant advancements have been made in image deblurring through multi-layer networks,but their performance remains limited by challenges in feature extraction and residual connections.To ad‑dress these issues,this paper proposes a multi-scale feature extraction and fusion network(MSFN)for image deblurring.The core idea of the network is to enhance image feature extraction through multi-scale inputs and outputs.Further,MSFN utilizes its feature adaptive detail enhancement(ADE)modules and cross-scale feature fusion(CSFF)modules to capture multi-scale features at different network depths,thereby optimizing the residual connection process and effectively integrating multi-scale information.Ex‑perimental results demonstrate that the proposed algorithm achieves superiority in quantitative analysis and significantly improves subjective visual effects,exhibiting an advanced performance.
作者
武婷婷
万少杰
WU Tingting;WAN Shaojie(College of Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China)
出处
《南京邮电大学学报(自然科学版)》
2025年第5期57-65,共9页
Journal of Nanjing University of Posts and Telecommunications(Natural Science Edition)
基金
国家自然科学基金(61971234)和江苏省研究生科研与实践创新计划项目(KYCX23_0960)资助项目。
关键词
图像去模糊
深度学习
多尺度
细节增强
特征融合
image deblurring
deep learning
multiple scale
detail enhancement
feature fusion
作者简介
武婷婷,女,博士,教授,博士生导师,wutt@njupt.edu.cn。
