NVIDIA Corporation (NASDAQ:NVDA) has teamed up with the researchers at Finland’s Aalto University and MIT to fix the grainy photos using artificial intelligence (AI) system. The team downloaded 50,000 photos from the dataset of ImageNet and trained its artificial intelligence system for reconstructing the photos.
Incidentally, the system named Noise2Noise also has the ability to eliminate noise from a particular image even though the system has not seen the image sans the noise. According to Jacob Munkberg, a Nvidia researcher, “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.”
Noise2Noise reconstructs clean image without any noise
The system has been created with the help of deep learning. Each of the 50,000 images drawn using the Noise2Noise AI system came as a high-quality, clean image without any noise. However, the team manipulated the high-quality image a bit to introduce randomized noise. The team also used MRI scans and computer-generated images for training the AI system Noise2Noise.
Noise reduction or denoising methods have been in use since long. However, the use of this method in tandem with deep learning started recently. The new method was revealed at the recently held International Conference on Machine Learning in Stockholm, Sweden. Interestingly, this method does not require a noise-free image for enabling AI to eliminate noise, grain, and artifacts and enhances the image automatically.
According to experts, the noise looks similar to a grainy snow in an image taken when the light is low. This noise can be found also in other forms of photographs such as computer-generated images, medical imagery and space pictures. More often, the images from digital cameras that are taken using the digital zoom or when the light is low have the noisy grainy snowy appearance.
THE HERALD FINANCE REPORT
Start your workday the right way with the news that matters most.
Researchers hope Noise2Noise can work on images with high amount of noise
Researchers have trained Noise2Noise only with noise and hence they hope that the method can be effectively used even on the images that comprise of a high level of noise like astrophotography, MRI, brain scans or the photographs that were robbed off clear definition as it was taken in the dusk.