非局部多尺度融合的单帧图像超分辨率重建  

Non-local multi-scale fusion for single image super-resolution reconstruction

作　　者：陈蓉[1] 李翠华[2] CHEN Rong;LI Cuihua(Information Engineering College,Xizang Minzu University,Xianyang 712082,China;School of Informatics,Xiamen University,Xiamen 361005,China)

机构地区：[1]西藏民族大学信息工程学院,陕西咸阳712082 [2]厦门大学信息学院,福建厦门361005

出　　处：《厦门大学学报（自然科学版）》2022年第2期278-287,共10页Journal of Xiamen University：Natural Science

基　　金：西藏自治区自然科学基金(XZ2018ZRG-65)。

摘　　要：现有的深度超分辨率重建模型,用堆叠多个相同模块的方式获取具有更高精度的重建结果,但未能充分考虑各层特征间的上下文关联信息.提出一种基于非局部多尺度融合的图像超分辨率重建模型.该模型采用3种模块:非局部模块、多尺度融合模块和宽激活残差模块.其中,非局部模块用于获取图像的全局特征,关注目标的核心区域;多尺度融合模块用于融合通道特征,增强特征在空间的上下文关联;宽激活残差模块替代普通残差模块,通过扩充激活层前的卷积层输出特征来提升模型的重建精度.在5个基准数据集上的实验结果表明,该模型取得了较高的重建精度.在图像分割与目标检测任务上,与几种深度超分辨率模型相比,该模型也取得了较好的分割和检测精度.In the existing deep super-resolution reconstruction model,stacking multiple identical modules to obtain higher-precision reconstruction results are used.However,the contextual information between features of each layer cannot be fully considered.In this study,we propose an image super-resolution reconstruction model based on non-local multi-scale fusion.In the model three modules are used,namely,non-local module,multi-scale fusion module and wide activation residual module.Among them,the non-local module is used so that global features of the image and focus on the core area of the target can be obtained.Next,the multi-scale fusion module is used to fuse channel features and enhance the contextual correlation of features in space.Finally,the wide activation residual module activates through expansion features before the Rectified Linear Unit to replace the ordinary residual module,which are used to improve the reconstruction accuracy of the model.Experimental results on five benchmark datasets show that the proposed model achieves high reconstruction accuracy.For the image segmentation and target detection tasks,compared with several deep super-resolution models,the proposed model has also achieved satisfactory segmentation and detection accuracy.

关 键 词：单帧图像超分辨率 深度超分辨率模型 非局部 多尺度融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]