Super-Resolution Image Reconstruction: Selective Milestones and Open Problems

In this talk, the presenter will review the development of image super-resolution technology based on the evolution of key insights associated with the prior knowledge or regularization method from analytical representations to data-driven deep models. The co-evolution of super-resolution with other technical fields, such as autoregressive modeling, sparse coding, and deep learning, will be highlighted in both model-based and learning-based approaches. Model-based super-resolution will include geometry-driven, sparsity-based, and gradient-profile priors; learning-based super-resolution will cover three types of neural network architectures, namely residual networks generative adversarial networks, and pre-trained models. Both model-based and learning-based SR are united by highlighting their limitations from the perspective of model-data mismatch. The presenter will also briefly discuss several open challenges, including arbitrary-ratio, reference-based, and domain-specific super-resolution.