This A. Cleans noisy images with one shot

Low-light images are often affected by grain, small dots created by increasing the sensitivity of the camera or ISO which obscure the finer details of the image. But researchers at Nvidia, Aalto University, and the Massachusetts Institute of Technology have formed a computer to eliminate the grain using only the original photo and software.

While the first artificial intelligence programs can clean up a noisy image, these programs required two photos, one full of grain and one without. Nvidia's new research, published on Monday, July 9, only needs a single grainy photo to create a sharper image using AI

The researchers formed the program by feeding the program. computer 50,000 pairs of images. The pairs were almost identical, except that each image in the pair had a different random pattern of grain added with software. Previous searches used pairs of images, but one image was a clean, low-noise file. The research, the group wrote, proves that it is possible to reduce the grain in an image without using a low-noise image as a reference point.

To test the program, the group used traditional images and even medical MRIs, suggesting the technology could be used for more than just cleaning low-light photos. The team used images with extra noise in order to have a proper reference image to see how the A.I performed. The resulting images had less noise than the original and took only milliseconds to correct. In the samples shared by the researchers, the AI-treated program was a little more flexible than the original reference image, but the adjusted images no longer had troublesome grain levels.

The researchers point out that the program can, of course, find details that are not there or have been obscured by the noise, but the program adjusts the images without a clear reference picture. "There are several real-world situations where it's difficult to get clean workout data: low-light photography (for example, astronomical imaging), physical rendering, and physical activity." Magnetic resonance imaging, "the researchers wrote. "Our proof of concept demonstrations pave the way for significant potential benefits in these applications by removing the need for a potentially laborious collection of clean data. Of course, there is no free meal – we can not learn to find features that are not listed in the input data – but this also applies to the same thing. training with clean targets. "

The research will be presented at the International Machine Learning Conference in Sweden later this week – but like most new research, there is still no word about if and when technology can be widely available.