On a decomposition model for optical flow

Abstract

In this paper we present a variational method for determining cartoon and texture components of the optical flow of a noisy image sequence. The method is realized by reformulating the optical flow problem first as a variational denoising problem for multi-channel data and then by applying decomposition methods. Thanks to the general formulation, several norms can be used for the decomposition. We study a decomposition for the optical flow into bounded variation and oscillating component in greater detail. Numerical examples demonstrate the capabilities of the proposed approach. © 2009 Springer.