Analysis and comparison of state-of-the-art methods in image super-resolution reconstruction

Image super-resolution reconstruction (SR) is one of the fundamental image processing techniques in computer vision, which not only improves the image resolution and quality but also assists other computer vision tasks. In recent years, with the rise of the artificial intelligence wave, image SR based on deep learning has made significant progress. This paper compares the technical architecture and research progress of image SR based on deep learning in recent years, including the summary of datasets, and the subjective and objective evaluation metrics used for image quality assessment. It focuses on analyzing methods proposed within 5 years, based on recurrent neural networks, generative adversarial networks, Transformers, pixel alignment, etc., and provides evaluations and comparisons of relevant network models. Finally, it presents an outlook on the future research trends of deep learning-based image SR algorithms.