摘要
近年来,伴随着人们对深度学习技术的不断钻研,人们可以越来越娴熟地使用深度学习实现图像超分辨率重建。针对多数网络训练时间长、重建高频细节缺失严重等问题,使用基于残差学习的图像超分辨率重建方法。网络通过残差学习的策略,加快网络的学习和收敛的速度,在残差网络的基础上,加入分数阶傅里叶变换最大限度恢复图片细节信息,将LR图像与使用网络学习得到残差图像相结合得出最终重建的HR图像。最终实验结果验证,与其他相关超分辨率重建算法相比,提出的方法得出的评价指标峰值信噪比(PSNR)与结构相似性(SSIM)值高于其他算法,重建效果更好一些。
In recent years,with the continuous study of deep learning technology,people can be more and more skilled in using deep learning to achieve image super-resolution reconstruction.For most network training time is long,rebuild the high frequency detail missing serious problems,this article USES the network through the residuals of learning strategy,speed up the network learning and convergence speed,on the basis of the residual network,join the fractional Fourier transform maximum recovery image details,the LR image study using neural network and the residual image combining the final reconstruction of HR The image.The final experimental results verify that the evaluation index PSNR and SSIM values obtained by the proposed method are higher than other algorithms,and the reconstruction effect is better.
作者
杨梦薇
肖秦琨
朱毅琳
Yang Mengwei;Xiao Qinkun;Zhu Yilin(School of Electronic Information Engineering,Xi'an Technology University,Xi'an 710016,China)
出处
《国外电子测量技术》
北大核心
2022年第4期170-175,共6页
Foreign Electronic Measurement Technology
基金
国家自然科学基金(62071366)项目资助
关键词
深度学习
图像超分辨率重建
残差学习
分数阶傅里叶变换
deep learning
image super-resolution reconstruction
residual learning
fraction-al Fourier transform
