Variational Approach for Joint Optic-Flow Computation and Video Restoration

We introduce a variational approach for simultaneous optical flow computation and video denoising. The proposed functional includes optical flow terms that depend on the restored sequence and an image sequence restoration term that depends on the optical flow. Our functional results in coupled Euler-Lagrange equations that are solved simultaneously for both the optical flow and the image sequence. The main novelty is the bidirectional coupling of the two problems. Traditional optical flow methods usually prefilter the sequence, an operation which does not depend on the computed flow. At the other end, in image restoration, most methodologies compute the optical flow as a preprocessing stage which is independent of the restoration procedure. Here, we formulate the dependency of the optical flow and image restoration problems for the denoising and deconvolution problems. Our experiments demonstrate that the new method achieves better optical flow estimation under substantial noise levels compared to previously reported results.