A survey of deep learning approaches to image restoration

In this paper, we present an extensive review on deep learning methods for image restoration tasks. Deep techniques, led by convolutional neural networks, have received a great deal of attention in almost all areas processing, especially classification. However, is fundamental and challenging topic plays significant roles understanding representation. It typically addresses deblurring, denoising, dehazing super-resolution. There are substantial differences the approaches mechanisms restoration. Discriminative based able to with issues mapping function effectively, while optimisation models can further enhance performance certain constraints. offer comparative study techniques dehazing, super-resolution, summarise principles involved these tasks from various supervised network architectures, residual or skip connection receptive field unsupervised autoencoder mechanisms. Image quality criteria also reviewed their assessed. Based our analysis, efficient deblurring couple multi-objective training functions super-resolution The proposed compared extensively state-of-the-art both quantitative qualitative analyses. Finally, point out potential challenges directions future research.