A model-based adaptive motion estimation scheme using Renyi's entropy for wireless video

Efficient motion estimation is an important problem because it determines the compression efficiency and complexity of a video encoder. Motion estimation can be formulated as an optimization problem; most motion estimation algorithms use mean squared error, sum of absolute differences or maximum a posteriori probability as the optimization criterion and apply search-based techniques (e.g., exhaustive search or three-step search) to find the optimum motion vector. However, most of these algorithms do not effectively utilize the knowledge gained in the search process for future search efforts and hence are computationally inefficient. This paper addresses this inefficiency problem by introducing an adaptive motion estimation scheme that substantially reduces computational complexity while yet providing comparable compression efficiency, as compared to existing fast-search methods. Our approach is motivated by the recent developments in using Renyi's entropy as the optimization criterion for system modeling [D. Erdogmus, J. Principe, Comparison of entropy and mean square error criteria in adaptive system training using higher order, in: Proceedings of the International Workshop on Independent Component Analysis and Signal Separation, Helsinki, Finland, 2000, pp. 75-80]. This scheme is particularly suited for wireless video sensor networks, video conferencing systems and live streaming videos which have stringent computational requirements. Our results show that our scheme reduces the computational complexity by a factor of 9-21, compared to the existing fast algorithms.