A parallel algorithm for side information generation in Distributed Video Coding

Most of the reported Distributed Video Coding (DVC) schemes have a high time-delay in decoder which hinders its practical application in real-time systems. In this paper, we propose a new DVC decoder architecture based on MPI clusters and an efficient parallel algorithm for generation of the Side Information (SI) based on the structure. In order to achieve a balanced workload among the processors and minimize the overheads of inter-process interaction, we adopt a task-pool model to implement dynamic mapping of tasks onto processors. In addition, the proposed method also uses a threshold-based approach to attain good trade-off between decoding time and video quality. We analyze the effect of computational granularity and the number of processors on performance through defining a performance function. The experimental results on multicore PCs show that a near-linear speedup can be obtained by using the proposed parallel SI generating algorithm.