A team from NVIDIA, the Massachusetts tehnologicheskogo Institute and Aalto University presented a neural network, is able to edit the image and remove the excess noise, as well as various inscriptions. On the work done by the specialists of NVIDIA shared in his official blog, but the source code for the algorithm not to spread in open access.
To create a neural network experts used a machine learning technique. On the basis of 50 different damaged photos, the algorithm has learned to remove image noise caused by low light, color noise old photos and get rid of labels, signatures, and perhaps even copyright, available on the images. The developers note that the neural network for effective work don’t need to see the same, but clean the photo — just to show poor quality images.
If shown in the APR algorithm NVIDIA has demonstrated the ability to restore to clean the photographs of persons missing parts, as well as the ability to remove the unnecessary details, the current algorithm is able to carry out all the work on the basis of “corrupted data”.
“You can teach a neural network to recover a portion of the photograph to the algorithm have not even seen its original value. When it does not suffer the performance of your computer, and does not increase the processing time”, — say the developers.
The developers note that the method can be used not only on treatment of the photographs but also to remove noise from MRI images, other medical images, as well as, when you receive the photos with low light level, for example, to perform astronomical observations.
NVIDIA has taught the neural network to remove unnecessary noise and inscriptions with photographs
Nikolai Khizhnyak