Trans-GAN network for image super-resolution reconstruction
Yan Dong, Xiaoyang Xu, Chong Gao · 2022 IEEE International Conference on Signal Processing, Communications and Computing (ICSPCC) · 2022
In order to solve the problems of network training difficulties and artifacts in the generated image super-resolution reconstruction algorithm based on convolutional neural network, this paper proposes a Trans-GAN super-resolution reconstruction algorithm based on generative adversarial network. By introducing Transformer model, the algorithm proposed a new Trans-GAN network, and improved residual module, using VGG19 network as the discriminator model framework, using adaptive pooling to replace the full connection layer to prevent overfitting, introducing per-pixel mean square error loss, perceived loss, The antagonism loss and total variation loss constitute the total objective loss function of the generator to solve the problem of insufficient feature extraction ability in the super-resolution algorithm based on convolutional neural network. Experimental results show that the proposed network can obtain better detail texture and image quality, and the PSNR of 4-fold image hyperfraction reconstruction is improved by 0.57 on average on five public datasets of Set5, Set14, BSD100, Urban100 and DIV2K.