摘要
为了改善低分辨率图像的视觉效果,增加图像细节信息量。文章结合现在比较流行的深度残差学习方法,提出一种新颖的残差网络结构,即深度多级残差网络。在残差网络中,当使用恒等映射作为捷径连接时,信号可以从一个单元直接传播到其他单元。基于此,该文在原有的残差网络结构上再加上多级捷径链接,挖掘残差网络的优化能力;针对不同的测试集,深层多级残差网络模型取得了更佳的超分辨率(super-resolution,SR)结果,不论是在主观视觉上还是在客观评价指标上均有明显改善,图像清晰度和边缘锐度明显提高。实验结果证明了深层多级残差网络对图像超分辨率重建的有效性,且网络的收敛速度更快,重建质量更好。
To improve the visual effect of low-resolution images and increase the amount of image detail information,this paper proposes a novel residual network structure called deep multilevel residual networks in combination with the more widespread deep residual learning method.In residual network,the signal can be directly propagated from one unit to any other units when using identity mapping as the skip connection.On this basis,level-wise connections upon original residual networks are added to dig the optimization ability of residual networks.For different test datasets,the model of deep multilevel residual networks achieves superior super-resolution(SR)results,and the subjective visual effect and objective evaluation indices are both improved obviously.The image resolution and edge sharpness are enhanced significantly.The experimental results demonstrate the effectiveness of the deep multilevel residual networks for image SR reconstruction,and the convergence speed of the network is faster and the reconstruction quality is better.
作者
吴从中
魏雪琦
詹曙
WU Congzhong;WEI Xueqi;ZHAN Shu(School of Computer and Information, Hefei University of Technology, Hefei 230601, China)
出处
《合肥工业大学学报(自然科学版)》
CAS
北大核心
2020年第10期1330-1336,共7页
Journal of Hefei University of Technology:Natural Science
基金
国家自然科学基金资助项目(61371156)。
关键词
图像超分辨率(SR)
深度学习
残差学习
卷积神经网络
捷径连接
image super-resolution(SR)
deep learning
residual learning
convolution neural network
shortcut connection