Learned Edge-Guided Multiple Attention-based Generative Adversarial Network for Single Image Super-Resolution

Abstract- It is well-known that recovering high-frequency details from low-resolution images remains a major challengein single image super-resolution. This work presents a multiple attention-based generative adversarial network thatleverages a learned edge detector to enhance structural details in the reconstructed images. A multiple attention-basedblock is designed to capture multi-level contextual information and different attention-based features. In addition, aScale-Invariant Feature Transform-based loss function is employed to ensure the preservation of scale-invariant featuresand reducing discrepancies between the generated and ground-truth images. Comprehensive experiments on standarddatasets demonstrate that the proposed approach consistently outperforms existing state-of-the-art models and achieveshigher PSNR and SSIM values while delivering perceptually superior results with lower LPIPS scores.