Dense Optical Flow Computation Using Textural Features

This paper proposes a new algorithm for dense optical flow computation. It combines Gabor filter texture description in a variational framework. The textural features with spatial Gabor filter group are used as an extension to the original bright constant model. The corresponding global energy functional is established, and the flow field at each pixel of the image is iteratively computed by solving the corresponding discretized Euler-Lagrange equations. Performances of the algorithm are tested on real and synthetic sequences, and are compared with other dense optical flow computation techniques, which demonstrate a great improvement on the convergence rate and robustness against brightness changes.