We show how optical flow estimates can be combined with boundary estimation to improve estimates of motion. The improvement is associated with blending of estimates from complementary bases of operation. The paper combines a phase-based method for optical flow with a time extended version of the phase congruency operator. By evaluation on synthetic and real image sequences, the combination of the two techniques is shown to improve motion estimation with particular advantages at motion boundaries, regions which have posed considerable difficulty for previous motion estimation techniques. The advantage is derived using the moving feature information in an extended phase congruency operator to constrain correct data in the optical flow field.