Deep Convolutional Neural Networks (CNN) have become a very popular tool for image creation and retrieval, with a number of new algorithms released in the last months of 2017 capable of achieving impressive results in reconstructing and enhancing damaged or low-resolution images. One of the most impressive algorithms is called Deep Image Prior, developed by a team of Russian scientists.
This algorithm stands out from the rest as it breaks away from the conventional methods. Instead of relying on large amounts of training data to choose the best way to handle the degraded image, Deep Image Prior uses data from the same image to reconstruct it to its original state.
The researchers report that the algorithm is capable of removing pixels from images, text from photos, gaps from images, smoothing out graphics problems due to aliasing of JPEG images, and even enhancing lower resolution images with acceptable results.
This ability to fix corrupted images is quite impressive, and it's the first time scientists have attempted to recreate Adobe 's Content Aware Fill/Brush tool, which was added to Photoshop a few years ago. A technological secret that no other image editing program has been able to replicate.
Deep Image Prior is one of two similar image processing projects developed this year. The second is called PixelNN . This one was developed by three researchers at Carnegie Mellon University and is capable of reconstructing blurry or partially corrupted images. Its difference from Deep Image Prior is that it needs to be “trained” with a large amount of training data, but it is much more accurate than other similar projects and is capable of recovering images with a high amount of corrupted data .
