The company Nvidia have developed technology based on artificial intelligence, which automatically restores damaged fragments of the image or fills its missing parts. The neural network algorithm understands the context of a photo or picture, and independently modifies it.
The method, called “painting an image,” is similar to the “Content-Aware Autofill” feature that Adobe added in Photoshop CS5 in the late 2000s. The technology also restored the deleted part of the image based on the background of the photo.
Previous approaches to in-depth training were focused on rectangular areas located around the center of the image, and often assumed costly post-processing. Our model can effectively handle areas of any shape, size or distance from the boundaries of the image.
In addition, at the end of 2017 a team of researchers from Yandex and Oxford University developed a similar algorithm called Deep Image Prior. Technology is a neural network that can restore damaged and low-quality images.
The algorithm from Nvidia can handle areas of various shapes and sizes. The researchers of the company stressed that they trained a neural network using more than 55,000 photomasks. They were used to remove some parts from the original images. Then the algorithm was shown the cut fragment and its original shape to teach it how to restore the missing pixels.
At the end of the tests, the developers confirmed the accuracy of their algorithm with the help of another 25 thousand photomasks. This time the neural network was not shown the original image, forcing it to restore the missing parts on its own.