Optical flow based super-resolution
An interesting task in computer vision is the super-resolution problem where, using the dynamics in an image sequence, images with increased spatial resolution can be obtained.
This can be used for example to enhance the resolution of telescopes or make a license plate in a video readable.
In this talk a variational framework for video super-resolution based on an optical flow model is proposed. We introduce a joint model that is simultaneously capable of calculating
high resolution images from all corresponding frames in a sequence and estimate the optical flow between them. Both tasks endorse each other and give a significant benefit to the
reconstructed image sequence and the obtained velocity fields.
During the talk we give a short overview of existing models, provide analytical results and give insight into an alternating minimization approach that leads to convex
sub problems. Finally, applications to synthetic and real data are shown.