Computational Aspects of Determining Optical Flow

We study some computational aspects of determining optical flow. Following the regularization method, we show the existence and the uniqueness of the discrete smoothing spine. To solve for the spline, Chebyshec iterative method is proposed, and its convergence and complexity is analyzed, with a comparison against Gauss-Seidel and Jacobi, which is prevalent in computer vision. Chebyshev method converges faster, and requires only simple local interactions. Testing results on synthetic images support the theoretical analysis.