A survey on Deep Learning-based Single Image Super-Resolution

Single Image Super-Resolution (SISR) is a fundamental task in computer vision with the aim to reconstructhigh-resolution images from low-resolution counterparts. Deep learning has revolutionized SISR in recent years, enablingsignificant improvements in both reconstruction fidelity and perceptual quality. This survey provides a comprehensivereview of deep learning-based SISR methods, categorizing them into four primary paradigms: convolutional neuralnetworks, generative adversarial networks, transformers, and diffusion-based models. For each category, thearchitectural designs, their strengths and limitations are analyzed. The challenges being faced in this area and potentialdirections for future have been discussed. This survey aims to provide researchers with a unified understanding of thecurrent state-of-the-art in deep-learning based SISR.