<p>Super-resolution (SR) aims to reconstruct high-resolution images from low-resolution inputs, with deep learning advancements driving substantial improvements in SR performance. This paper presents a comprehensive review of single- and multi-image SR techniques, analyzing findings from 12,873 research papers published between 2015 and 2025 in the computer science field. Key insights are derived from fifteen summary tables covering various SR tasks, including natural, medical, video, burst, depth map, and underwater image SR. The analysis highlights several major findings: (1) the integration of specialized modules, such as attention mechanisms, has led to consistent yearly improvements in performance metrics like PSNR and SSIM; (2) domain-specific architectures often outperform general models, particularly in medical and underwater SR applications; (3) while benchmark datasets enable objective comparisons, real-world validation remains limited, reducing the generalizability of current approaches; (4) inconsistent metric reporting across studies hampers reproducibility and fair evaluation; and (5) practical deployment considerations, including computational efficiency and real-time processing, are rarely addressed. Despite significant progress, challenges such as the need for more diverse training datasets, robust validation, and better interpretability persist. This review synthesizes these critical findings, offering an updated perspective on SR advancements, emerging trends, and future research directions.</p>