Super Resolution Optical Flow

Abstract

Existing approaches to super-resolution are not applicable to videos of faces because faces are non-planar, non-rigid, non-lambertian, and are subject to self occlusion. We present super-resolution optical flow as a solution to these problems. Super-resolution optical flow takes as input a conventional video stream, and simultaneously computes both optical flow and a super-resolution version of the entire video. In this paper we describe an initial implementation of super-resolution optical flow, present detailed experimental results, and describe the relationship between super-resolution optical flow and pyramid-based image representations such as the Laplacian pyramid.