The iterative graph cut optical flow estimation with space term disturbance

This paper proposes a novel iterative α -expansion move optical flow estimation with space term disturbance to decrease the local error caused by outlier points. At first, the reason for less smoothness of the discrete optical flow estimation is analyzed. And then, the truncated linear function is adopted and inhibits outlier points caused by the noise and the occlusion because its disturbance makes the solution jump out of the local minimum. Given the theory of the graph cut, the property of decreasing further the energy by disturbing the space term is proven. The experiment shows that this approach could reduce outlier points and smooth further the optical flow field than the estimation with fixed space term.