Block Motion Estimation and Potential Games

摘要

In this paper, the problem of inter-prediction block motion estimation (ME) is cast in a game theoretic framework. We model the optimization problem of ME using a network of players in a game theoretic approach. First, a global objective function that captures the notion of consensus is established. Next, it is shown that local objective functions can be assigned to a network of players, so that the resulting game is proven to be a potential game. This game can then be solved in a distributed manner where each player optimizes its own utility function. Consensus strategies can be employed to allow all the players to reach the common minimizer of the global objective function. The resulting scheme is an accurate and highly parallel ME algorithm that decreases the computational burden of the full-search scheme while preserving the quality of the produced motion information.