RAISR

from Wikipedia, the free encyclopedia

RAISR (Abbreviation for Rapid and Accurate Image Super Resolution , in German about fast and accurate image super resolution ) is an algorithm from Google to enlarge photos. The software that uses machine learning has been in use since the end of 2016.

RAISR is a method to compute high-quality images from low-resolution versions. It delivers results that are qualitatively comparable to or better than existing super-resolution methods and shortens the calculation time by 10 to 100 times. The speed of the process enables the images to be made available on mobile devices in real time. RAISR also avoids restoring aliases that are still present in the lower-resolution versions. With a quality comparable to the original, the algorithm can save around 75% of the bandwidth.

RAISR uses filters that are selectively applied to individual areas of the low-resolution image in order to restore details. These filters are trained in RAISR using deep learning , in which the program compares a low-resolution and a high-resolution version of an image. The filters were learned from variants of different resolutions as well as from an image initially enlarged by conventional upsampling methods with the high-quality image. RAISR received additional training for edge detection .

The process is mainly used to upscale images on the client side on mobile devices and thus to save bandwidth when transmitting the images. In December 2016, RAISR was first used in the iOS app Motion Stills , an app that uses image stabilization to create GIFs . The process has been in use at Google+ since 2017 , where one billion images were scaled every week in January 2017. The main camera and the pixel Visual Core of pixels 2 Google use RAISR to have to digitally zoomed images look sharper and more detailed.

Web links

  • Peyman Milanfar: Enhance! RAISR Sharp Images with Machine Learning . In: Research Blog . November 14, 2016 ( googleblog.com ).
  • Yaniv Romano, John Isidoro, Peyman Milanfar: RAISR: Rapid and Accurate Image Super Resolution . In: arXiv.org . June 3, 2016, arxiv : 1606.01299 [abs] .

Individual evidence

  1. ^ Matthias Grundmann, Ken Conley: Get moving with the new Motion Stills . In: Research Blog . December 15, 2016 ( googleblog.com [accessed February 2, 2017]).
  2. ^ John Nack: Saving you bandwidth through machine learning . In: Google . January 11, 2017 ( blog.google [accessed February 2, 2017]).
  3. ↑ Ofer Shacham: Use Pixel 2 for better photos in Instagram, WhatsApp and Snapchat . In: Google . February 5, 2018 ( blog.google [accessed May 7, 2018]).