In computer vision, image super-resolution is an essential task of image processing for enhancing the resolution of images from low to high resolution. Over the past years, many research efforts have been conducted to advance the performance of super-resolution algorithms.

Image super-resolution has a broad range of real-world applications, such as medical imaging, surveillance, text image enhancement, biometrics, remote sensing, security, and media. Deep learning has recently gained tremendous success in resolving the problems in image and video super-resolution problems. Advances in super image resolution are made with the help of deep learning. Image super-resolution methods comprise single-image super-resolution and multiple-image resolution methods. Interpolation, reconstruction, and regularization are some of the image-resolution techniques.

Furthermore, deep learning models are effectively applied for real-time image super-resolution methods. Convolutional Neural Networks and Generative Adversarial Networks have gained more attention to conduct deep learning based on real-time image super-resolution. Some top model frameworks are pre-upsampling, post-upsampling, progressive upsampling, iterative up-and-down, and sampling. Some recent improvements integrated into super image resolution are Context-wise Networks, Fusion, Data Augmentation, Multi-task Learning, Network Interpolation, and Self-ensemble. Potential concepts in deep learning-based image super-resolution are;

•  Supervised Image Super-resolution: Network architectures in supervised image super-resolution are Residual Learning, Recursive Learning, Multi-path Learning, Dense Connections, Attention Mechanism, Advanced Convolution, Region-recursive Learning, Pyramid Pooling, Wavelet Transformation, xUnit, and Desubpixel. Learning Strategies of supervised image super-resolution are Loss Functions, Batch Normalization, Curriculum Learning, and Multi-supervision

•  Unsupervised Image Super-resolution: Unsupervised image super-resolution method includes Zero-shot Super-resolution, Weakly-supervised Super-resolution, Learned Degradation, Cycle-in-cycle Super-resolution, and Deep Image Prior.

Domain-specific applications of deep learning-enabled image super-resolution are highlighted below:

•  Depth Map Super-resolution - Depth map super-resolution helps to increase the spatial resolution of depth maps in many real-time tasks such as pose estimation and semantic segmentation. Recently, CNN models have been utilized.

•  Face Image Super-resolution - Face hallucination is a promising research concept to assist face-related tasks by incorporating face-prior knowledge

•  Hyperspectral Image Super-resolution - Hyperspectral image super-resolution helps to obtain high-quality hyperspectral images by combining high-resolution and low-resolution panchromatic images.

•  Real-world Image Super-resolution - Real-world image super-resolution is an image restoration problem that focuses on obtaining high-quality images, and recently, it has gained sustainable attention for its remarkable application potential.

•  Video Super-resolution - Video super-resolution processes multiple frames in intra-frame spatial and inter-frame temporal dependency by using explicit motion compensation, convolutional methods, and recurrent methods.