A distributed adaptive block matching algorithm: Dis-ABMA

摘要

In the context of motion estimation for video sequences processing, variable block size algorithms, like the Adaptive Block Matching Algorithm (ABMA), have been proposed to match better “objects in motion” compared to the classical BMA. However, the variable block size grid derivation and the related motion estimation relies on a regularization process which implies heavy iterative and inter-dependent computations. Though the parallelization of the BMA is straightforward, the ABMA needs a deeper analysis before its implementation in a distributed environment: this is the goal of this paper. We first designed a modelization of the motion estimation of ABMA. This model can lead to several different distributed versions. A specific distributed model, with one master and several slaves, is then described. An implementation of this model has been realized and experimentations demonstrate a linear speedup with respect to the number of processors.