Superresolution videos and optical flow via combinatorial and variational optimization

Grant BR 3815/7-1

Project members

Prof. Dr. Thomas Brox
Prof. Dr. Daniel Cremers (TU München)
Peter Ochs
Eddy Ilg


Humans can draw very precise information even from videos of very bad quality. This astonishing capability gets evident as we regard single images of a video and realize how noisy and blurred they typically are. This is still true in the age of HD videos, which formally have a high resolution, but due to natural limitations during recording cannot provide the same quality as a single static image. In this project, we want to combine the information of successive frames in a video, such that the quality and resolution of all single frames can be increased. We assume that optical flow estimation, denoising, and superresolution are coupled problems. Based on the substantial preliminary works on these problems, we intend to develop a methodology, which jointly computes precise optical flow and high resolution images. Parts of this project will model motion blur, fast motion, and occlusion in the context of video-superresolution. We believe that the coupled optimization will help us with video analysis, the restauration of old movies, and the transformation of movies with limited resolution to full HD resolution and beyond.


IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Jul 2017

Alexey Dosovitskiy, Philipp Fischer, Eddy Ilg, P. Häusser, C. Hazırbaş, V. Golkov, P. Smagt, D. Cremers, Thomas Brox
IEEE International Conference on Computer Vision (ICCV), 2015

Peter Ochs, Y. Chen, Thomas Brox, T. Pock
SIAM Journal on Imaging Sciences, 7(2):1388-1419, 2014

IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2013