Parallel Adaptive Spatial-Temporal Prediction with Load Balancing for Fast Video Compression

Abstract

Video image compression has been an area where the computational demand is far above the capacity of conventional sequential processing. This paper presents a parallel adaptive prediction model for video compression using cluster computing on a local network with balanced load. The method used for prediction is adaptive, in which frames with very few motion changes are predicted in their temporal domain using motion estimation and frames with high motion activity are predicted in their spatial domain. This approach gives a good compression rate. A parallel compression model is applied by using multiple networked heterogeneous personal computer systems that perform compression on different input frames simultaneously. Also computing load is distributed properly among all the processors by the resource management technique of cluster computing. The experimental results show that the proposed parallel method has better speedup than the sequential algorithm and is suitable for fast compression of video data.

Keywords: Data parallelism, Task parallelism, Motion vector, Temporal predictor, Spatial predictor, Message passing interface