Glocal-Focused Super-Resolution Generative Adversarial Network

The rise of generative models has significantly propelled the development of Single - Image Super - Resolution (SISR) technology. However, key challenges still remain, including the loss of high - frequency details and high computational costs. In this study, we propose a novel Transformer - based generative adversarial network for super - resolution reconstruction, named SR - TransGAN. Our approach utilizes the Transformer as the core component of both the generator and the discriminator. It integrates a global feature module to capture in - depth global feature information, and a local feature extraction module with spatial and channel attention mechanisms embedded to capture local feature information. Our design effectively reduces the computational complexity. Additionally, we optimize the loss function of the original model, thus enhancing the super - resolution performance. Experimental results demonstrate that compared with existing methods, SR - TransGAN performs exceptionally well in terms of detailed textures and overall image quality. When the magnification factors are $\times 2, \times 3$, and $\times 4$, this method achieves significantly better results in terms of the Peak Signal - to - Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) metrics, highlighting its effectiveness.