Improvement of Local Minima Problem and Its Effectiveness for Motion Vector Detection Using Steepest Descent Method

In this paper, we propose a fast motion vector detection algorithm using steepest descent method which is one of gradient methods. Proposed method is capable of avoiding the local minima problem by employing adaptive initial vector setting and hierarchical algorithm, and obtaining improvement of prediction performance. From simulation results, we compare prediction performance and computational complexity with other methods, and show effectiveness of the proposed method.