Researchers such as Nvidia have successfully used AI to remove noise from photos and look forward to its application in the medical field.

Researchers at Nvidia, the Massachusetts Institute of Technology and the University of Arut are using artificial intelligence to eliminate noise in photos. The team used 50, 000 images from the ImageNet database to implement AI system training for photo reconstruction. Although the system has never seen a noise-free image, the system successfully removes the noise from the image.

The AI system, named Noise2 Noise, is built with deep learning and draws intelligence from 50, 000 images in the ImageNet database. The images are clean, high image quality, no noise, but add random noise. Computer generated images and MRI scans are also used for Noise2 Noise training.

Although the method of removing and correcting noise has been used for many years, the active use of deep learning has attracted people's attention, which is a recent thing.

The noise may look like gray snow in photos taken in low light, and can also be seen in medical images, computer-generated images and images of the universe. Digital camera images are also often included in images taken with insufficient light or using zoom.

By training Noise2 Noise only with noise, the researchers hope that this method can also be used in pictures of celestial bodies taken in the dark, MRI, brain scans and other images that are known for their high noise.

IXI data sets nearly 5000 images of nearly 50 places of the human body, which are used for MRI intelligence training of Noise 2 Noise. Although something that looks blurred than the original image without artificial noise is used, clarity seems to have been restored.

Nvidia researcher Jacob Munkberg commented on the following in a telephone interview from VentureBeat:

"this proves the concept that we have been using the open MRI database for training. At some point in the future, we may also be able to apply it in the field. "

The results of the study will be presented this week at International Conference on Machine Learning in Stockholm, Sweden.

[via VentureBeat] @ VentureBeat [original]

BRIDGE Members