DISTRIBUTED PARALLEL COMPUTATION IN VIDEO CODING USING WORKSTATIONS

Abstract

In this paper a new approach for coding moving pictures is presented. Because of the large number of calculations, the conventional solution uses tightly coupled multiproces- sors working in parallel to achieve real-time processing (encoding 25 - 30 pictures per second). A new idea is to distribute the workload among workstations connected to a network where a software package (e.g. PVM - Parallel Virtual Machine) supports the communication between the machines. In contrast with the present hard wired structures, this loosely coupled system provides more flexibility in coding algorithms and has better cost/performance. The paper describes the main parallel structures already used in video processing, and discusses the possibility of mapping them to this new paralell system. Also. simulations were carried out to examine the performance of the most computation- ally intensive operations (DCT - Discrete Cosine Transform and motion estimation). The tests were performed on a cluster of SUN Sparc 2s connected via Ethernel. It was experienced that DCT did not show any speed-up because of the extremely low CC ra- tio. However, motion estimation worked well if either a full or hierarchical search was used. This research work was carried out in 1994 at the Information Theory Group of the Department of Electrical Engineering, Technical University of Delft.