Deep Image Prior is an interesting technology for improving resolution and fixing images that have been damaged, for example, by having text superimposed, high levels of noise or corrupted. The technique uses neural networks, but unlike some other approaches, does not need training or pre-training. The research is being performed at Oxford University and in Russia.
Applications including de-noising, the removal of JPEG artefacts (yes, please! – Man. Ed.), super resolution, sharpening and inpainting by fixing the missing parts of images.
This Deep Image dejpeging process took around 3,000 iterations.
