PhD Project Kristof Denolf
Coding and Complexity Efficiency Relationship of Hybrid MC/DCT Video Codecs
1. INTRODUCTION
During the complete history of video coding, removing the redundant information
in video data while still preserving the best possible quality has always
been the primary goal: the search for the highest coding efficiency. The
work of the Joint Video Team (a collaboration of ISO/MPEG and ITU/VCEG)
to standardize a new generation video coding proves this search continues
today. During the last years, the implementation cost of these compression
systems became more and more important to avoid falling into the trap of
developing an extremely efficient coding algorithm never able to fit into
a realistic design. Moreover, the large complexity variation between average
usage cases and the most demanding scenario excludes worst-case implementations
and hence requires compromises.
Prior art in Network Quality of Service (QoS) typically deals with the
bandwidth versus quality trade-off. This PhD project adds an additional
dimension: the impact of limited processing resources on the quality (Terminal
QoS). Hence, the processing load will be combined with the coding efficiency
to analyse the relation of the bitrate, the quality and the complexity
of a hybrid Motion Compensated/Discrete Cosine Transform (MC/DCT) video
codec. The goal is not to improve the compression efficiency, but rather
interpret the efficiency of existing codecs in a broader and more complete
way by including complexity issues. More specifically, the focus will be
on MPEG-4 video, including the new part 10: the Avanced Video Codec (AVC).
2. VIDEO EFFICIENCY DIMENSIONS
A solid study of the bitrate/quality/complexity cost relations requires
first their proper definition. For none of them this definition is trivial.
Quality, complexity and even bitrate can each individually, only be faithfully
characterised in a multi-dimensional space to allow for generality and
completeness. Weighting and combining multiple dimensions of one component
for simplified analysis, reduces the level of detail, but at the same time
requires a thorough understanding of the goals to achieve, i.e. a meaningful
compromise between the cost axes. Eventually, a three dimensional quality\complexity\bitrate
plot is instantiated for a particular implementation platform and known
network parameters.
3. SCALING HYBRID VIDEO CODECS
The proliferation of video coding applications and the diversity in bandwidth
and processing resources of content consumers steer the analysis of the
quality, bitrate and complexity when scaling a video codec by varying the
quantization degree, the temporal and spatial resolution. From the cloud
of operation points, a pareto surface with optimal codec settings can be
extracted. The specific issues related to different usage scenarios require
two distinct approaches in this study: first a point-to-point connection,
then extend to a multi-user environment.