CS631 Paper Response Form
Paper Title: Performance of a Software MPEG Video Decoder
Author(s): K. Patel, B.C. Smith and L.A. Rowe
Main Point(s):
The paper describes techniques for optimizing a portable software MPEG decoder. Performance evaluation showed that memory bandwidth is the main performance bottleneck and
not the computation of the IDCT, as commonly assumed. Besides describing coding optimization, the
paper also explores and compares different dithering techniques, since dithering is together with
reconstruction the most memory intensive part of an MPEG decoder. The authors also propose a metric for the evaluation of MPEGs with different bit rates across platforms that relates the percentage of required bit rate to the price of the hardware. The MPEG decoder runs at about 2 times real-time for 320x240 video on an HP 750.
Possible Use(s):
The described MPEG player is public domain and thereby facilitates video-related research.
The experiences and of the authors and their optimization techniques are valuable for
various strands of video processing research, since performance issues are always very
salient in this domain.
Extensions:
The evaluation showed that the MPEG decoder is not fast enough on most of the current machines.
In order to make it run in real-time, one would have to make compromises in temporal and/or
spatial resolution and the dithering quality. That is, a useful extension would be to
incorporate the MPEG player into a tool for real-time playback that allows to adjust the
trade-off to user preferences and the particular application it is used for.
CS631 home page