A team of researchers at Oxford University in the United Kingdom and Skolkovo Institute of Science and Technology in Moscow, capital of the Russian Federation, have developed a neural network technology capable of rendering low-res and corrupted images into clear pictures.
The system, called Deep Image Prior, uses the reverse process of image learning based neural networks. Rather than learn through reviewing tons of related images, Deep Image Prior works within the corrupted or low-res image to shape the contours and sharpen the image using information within only that image.
The result is quite impressive, to say the least.
Co-author of the research Dmitry Ulyanov explained that the “‘network kind of fills the corrupted regions with textures from nearby.’ Instead of using the data from the datasets, Deep Image Prior redraws a blurry or damager picture until it gets it right. According to Interesting Engineering, some images turn out to be even better than the original input.”
Though the initial results are impressive, it is not perfect: “The obvious failure case would be anything related to semantic inpainting, e.g. in-paint a region where you expect to be an eye — our method knows nothing about face semantics and will fill the corrupted region with some textures.”
The team sees the technology having applications for everyday photographers as well as museums and archives that may need to restore old, grainy photos. One concern remains, however, and that concern centers around the hot topic of copyright infringement. Deep Image Prior tech can also be used to remove the text placed over images, making image theft easier than ever.
Another concern that DIY Photography raises is if this technology could one day replace photo restorers, a fine art that is honed after years of training. The primary problem outside of displacing this profession is whether or not a computer AI could do a better job at restoring photos than the human eye, a story that only time will tell.