摘要
针对视频超分辨重建过程中出现的噪声放大、特征丢失等问题,在BasicVSR++模型的基础上提出了一种基于注意力与反注意力机制的传播模型来对视频进行超分辨率重建处理.模型将原有特征分解为传播特征与冗余特征.传播特征在传播网络中传递信息,而冗余特征则在残差网络进行深度提取,最后PixelShuffle网络将得到的两部分特征进行融合和重建,得到了更好的超分辨率重建结果.在公开的REDS数据集中,评估指标PNSR(峰值信噪比)达到32.48dB,视频超分辨重建性能得到提升.
Aiming at the problems of noise amplification and feature loss in the process of video super-resolution reconstruction,on the basis of BasicVSR++model,a propagation model based on attention and reverse attention mechanism is proposed to optimize the video super-resolution reconstruction.The model decomposes the original features into propagated features and redundant features.The propagation features spread the information in the propagation network,while the redundancy features are extracted in depth at the residual network,and finally,the PixelShuffle network fuses and reconstructs the acquired two parts of the features to achieve better super-resolution reconstruction results.In the published dataset REDS,PNSR(Peak Signal-to-Noise Ratio)reaches 32.48 dB,the performance of video super-resolution reconstruction is improved.
作者
谢思宇
周斌
胡波
XIE Siyu;ZHOU Bin;HU Bo(South-Central Minzu University,College of Computer Science,Wuhan 430074,China;South-Central Minzu University,Key Laboratory of Information Physics Fusion and Intelligent Computing of the National Ethnic Affairs Commission,Wuhan 430074,China;Wuhan Dongxin Tongbang Information Technology Company,Wuhan 430074,China)
出处
《中南民族大学学报（自然科学版）》
CAS
2024年第4期504-512,共9页
Journal of South-Central University for Nationalities：Natural Science Edition
基金
湖北省技术创新专项基金资助项目(2019ADC071)
中央高校基本科研业务费专项资金资助项目(CZY23006)。
关键词
视频超分辨重建
注意力与反注意力机制
传播网络
video super-resolution reconstruction
attention and reverse attention mechanism
propagation network
