基于周期激活函数的单图像超分辨率深度残差网络  

Deep Residual Networks for Single Image Super-Resolution with Periodic Activation Function

作　　者：王梦露 WANG Menglu(School of Mathematics,Hefei University of Technology,HeFei 230601,China)

机构地区：[1]合肥工业大学数学学院,合肥230601

出　　处：《大学数学》2021年第6期11-15,共5页College Mathematics

基　　金：基于细分的几何逼近与可视媒体处理中的非线性理论与方法(61472466)。

摘　　要：提出了一种具有周期激活函数的深单超分辨残差网络.因为周期函数具有处处可微性,可以快速地适应复杂信号,从而将自然图像空间更精确地参数化,并以连续的方式表达图像,减少了超分辨过程中细节的损失,提出的模型,通过降低深度模型的冗余性,融合创新的激活函数,重建的超分辨率图像质量在Peak Signal to Noise Ratio(PSNR)与Structural Similarity(SSIM)标准下均有所提升.A deep single super-resolution residual network with periodic activation function is proposed.Due to the differentiability of the periodic function,the periodic function can adapt to the complex signal quickly,so as to more accurately parameterize the natural image space and express the image in a continuous way,which reduces the loss of detail in the process of super-resolution.The proposed model reduces the redundancy of the depth model and fuses the innovative activation function.The quality of the reconstructed super-resolution image is improved under the Peak Signal to Noise Ratio(PSNR)and Structural Similarity(SSIM)standard.

关 键 词：周期函数 卷积神经网络 深度残差网络 图像超分辨率 

分 类 号：O244[理学—计算数学]