Residual Net Use on FSRCNN for Image Super-Resolution
Jinlu Zhang, Mingliang Liu, Xiaohang Wang, Chengcheng Cao · 2021
In this paper, we present a new method based on Accelerating the Super-Resolution Convolutional Neural Network (FSRCNN) and residual network for the aim of image super-resolution. In our works, we fuse the network structure of FSRCNN and skip connections as the first step of our method, this part can transfer the original features to the training results effectively, in addition, we also change the deconvolution layer to a sub-pixel convolution layer. Our experimental results show that our method is easy to converge in training works and the image reconstruction result also has better quality. The proposed method has better effects corresponding to the reconstruct images on the Set5, Setl4 andGeneral-100 data sets, its average PSNR can increase about 0.38dB.