Multi-Scale Deep Neural Networks for Efficient and High-Quality Image Super-Resolution: A Review

Abstract

Image super-resolution (SR) has become a pivotal task in computer vision, driven by applications across fields like medical imaging, remote sensing, entertainment, and scien- tific research. With the evolution of deep learning, multi-scale deep neural networks have emerged as a powerful approach for enhancing image quality while maintaining computational efficiency. This review provides a comprehensive survey of multi- scale architectures in deep learning-based SR. We analyze key models, highlight their innovations, compare performance on benchmark datasets, and discuss current challenges and fu- ture research directions. By integrating multi-scale strategies, SR models achieve superior reconstruction accuracy, efficient processing, and better generalization across diverse conditions, paving the way for robust real-world deployments.