Noise is the most annoying distortion when taking pictures.
It can appear as grainy snow in photos taken in low light. It can also be present in other forms of photos, like medical imagery, computer-generated images, and pictures of space. Photos taken by digital camera photos in low light condition or photos taken using a digital zoom can also contain noise.
While it's possible to fix these noisy images to some degree with software and expertise, but an AI developed by Nvidia and MIT collaborating with Aalto University can do this much better. And that without even seeing the image without the noise first.
The AI is called 'Noise2Noise', and it has been trained using 50,000 pictures from the ImageNet dataset that include 5,000 MRI scans from 50 human subjects and computer-generated images that had randomized noise added to them.
In their paper, the researchers have shown that their AI can successfully remove enough noise to make the pictures usable again, with details and clarity that are very close to the original images.
From the results shown above, the images appear a bit more blurry than the original image without artificial noise. But still, they appear to have restored most of the details.
"This is a proof of concept that we trained on a public MRI database, but it might show promise sometime in the future that this can be practically applied," said Nvidia researcher Jacob Munkberg.
Using this advancement, the researchers hope that the method can be used on images known to contain high amounts of noise.
The AI is somehow similar to the neural network developed by researchers from Intel and the University of Illinois at Urbana-Champaign, which brings details to extremely dark images. However, Noise2Noise should perform better when it comes to real-time image processing and denoising, as the AI's deep learning algorithms were trained only with noise.
Denoising or noise reduction methods have been around for a long time, including ways to enhance low-quality images. While most of the time, the process involves manual labor that can be quite extensive and time consuming, using AI to do the job has become a more recent phenomenon.
Examples include:
An AI from Google that can unblur pictures, and also one AI from Tencent that can unblur photos. Then there is an AI that can reconstruct partially damaged pictures, an AI from Max Planck Institute for Intelligent Systems that can also enhance low-resolution photos. Then an AI from Let's Enhance that uses neural networks to upscale photos.