Software-based video encoding using high-performance computing

Abstract

Emerging multimedia applications demand faster and more flexible digital video coding. While hardware-based implementations are usually fast, the flexibility offered by software and the growing speed of general-purpose processors make software solutions promising alternatives. However, the immense computational requirement of video coding presents a strenuous challenge to software-based implementations. The contributions of this work are two fold. First, we exploit parallel and distributed computing to offer a plausible software solution, especially for higher resolution MPEG-l-based video in both on-line and off-line environments. For on-line applications, we exploit fine-grained data-parallelism within each frame of video, using a macroblock as the basic data unit. For off-line applications, we use coarse-grained data partitioning with a group of frames as an independent data unit. We propose efficient scheduling of I/O operations with a view to removing the I/O contention. We employ various parallel and distributed platforms ranging from supercomputers to clusters of workstations connected via fast interconnection networks working as virtual parallel machines. Our second contribution is a single-processor H.263 based coding solution. For low-bit-rate applications, real-time encoding and decoding using only a single processor PC or workstation is highly desirable. We design fast algorithms for major computation-intensive functional modules, and devise ways to use low-level machine primitives in such modules, to speedup their computation. The proposed solutions efficiently utilize available resources in order to achieve maximum encoding performance.

Date of Award	1999
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology