Feature Assisted Fast Motion Estimation Algorithm

Block motion estimation is one of the most significant and computationally demanding component in video encoding. Most of the search algorithms start their search from the center of the search window and speed up the process by limiting the search locations. The search however, may get trapped in a local minimum on the error surface resulting in suboptimal quality. Another drawback is that the blocks located on the edges of moving objects are not well estimated, to which the human visual system is very sensitive. The technique proposed in this paper employs features in a macroblock to place the initial search point near to the global minimum of the error surface. Consequently, a fast and more efficient motion estimation algorithm is developed. The computational feasibility, better objective quality and the fact that it works as a simple extension to the existing search algorithms are the main advantages of the algorithm.