Putin signs law on ratification of Russia air force grouping in SyriaRussian Politics & Diplomacy July 27, 9:49
Moscow clarifies its stance on new anti-Russian sanctions to BrusselsRussian Politics & Diplomacy July 27, 9:35
Russia’s Defense Ministry launches Chinese and Arabic versions of 2017 Army Games websiteMilitary & Defense July 27, 8:55
Finland succeeds in building prosperous state for 100 years of independence — presidentWorld July 27, 8:27
Russian expedition recovers unique naval gun from Kerch StraitSociety & Culture July 27, 8:18
Brussels likely to declare US anti-Russian sanctions invalid, Moscow's EU envoy saysRussian Politics & Diplomacy July 27, 3:41
Russian PM calls situation around Saakashvili's citizenship a weird tragicomedyRussian Politics & Diplomacy July 27, 3:36
Russian Ice Hockey Federation to wage ruthless war on doping abuseSport July 26, 19:53
Two Siberian residents jailed for killing three zoo birds in failed barbeque attemptSociety & Culture July 26, 18:43
MOSCOW, May 31. /TASS/. Researchers from the Skoltech Institute of Science and Technology have unveiled a new unrivaled innovation for reducing noise in digital images based on the application of a neural network, Skoltech’s press office said.
Noise-reduction is crucial for working with digital images, as well as for computer vision. The task of removing noise consists of eliminating visual distortions while keeping the distinct uniqueness of an image, such as edges or texture.
"In our work, by making use of the latest achievements in deep learning, we created a new neural network which can greatly eliminate any defects from earlier innovations for reducing noise on images, which brings about favorable results in a relatively short period of time," said Stamatios Lefkimmiatis, professor at Skoltech, and leader of the research group who came up with the new approach.
The newly created technique is based on algorithms of deep learning of neural networks. The scientific team put together a neural network which effectively releases black-white and color images from noise. The network also takes into account the self-similarity properties of images which usually contain repeating sections. In contrast to previously existing technologies, the neural network considers this informational redundancy which is beneficial to the quality of pictures’ outcome.
According to the scientists, neural network learning might take quite a long time. Nevertheless, after the learning process is completed, the smart network can provide reasonable results relatively quickly.