Grainy photos due to low lighting are eternally sin. Short shutter speeds and a small aperture can also result in grainy photographs. You can throw these photos away or edit them into something that ultimately does not produce a beautiful photo. Researchers have devised a system with AI to get the noise out of the photos.
Artificial intelligence and photos
Researchers from Nvidia, Aalto University and MIT have trained a computer to make up for a grainy picture. All that is needed is the original photo and software.
Artificial intelligence existed which could improve such photographs, but this required a grainy and a non-grainy photograph. The AI of Nvidia only needs a grainy photo to make a pure photo afterwards.
Training the computer with input
Before a computer itself can do anything, it has to be trained with input. The researchers have input 50,000 pairs of images as input into the computer. Each pair was the same photo, but with a different pattern of graininess, adjusted with software. The original photo then served as a comparison material. No low-purity photos were used, so they were all grainy. The computer never saw a photo with little noise.
The photos to train the computer consisted of real photos, but also MRI scans. The photos that have been clarified by the AI came very close to the pure photo. The computer only needed a few milliseconds to adjust the photo.
The most interesting thing about this research is that with two grainy photographs a pure photo can be made, without needing a pure photo. This does not mean that the computer can adjust details that are not there, but that it can make an impure picture clearer.
The application can be used to make normal photos less grainy, but it can also contribute in the medical field. MRI scans can be clarified by making the scans less granular.