I am not much given to the expression of disclaimers, but here
goes.
Disclaimer
- The Waterloo BragZone is not a contest. Even though I refer to
a program as a "contender" or "entrant," there are no awards,
ribbons, or prizes. This is not an attempt to find the "best"
compression program in existence. Rather, it is a gathering of
information. Results are presented from a wide range of programs
to establish watermarks of performance.
- Take case when comparing. A compression program can be a
complicated thing, comprised of many elements. The vastly different
results of fractal and DCT coders, for example, are due to a
vast number of design differences -- no one thing can be isolated
as the cause. Only in a few cases do tests vary according to one
variable only. (Two examples: the difference between JPEG and
IJPG is the flavor Huffman coding; the difference between TRNA
and TRNB is the depth of quadtree used.)
- Beware of generalizing from the data. For instance, that the class
of wavelet coders are inherently superior to DCT coders. Personally
I believe this is true, but establishing the proposition requires a
carefully reasoned analysis. Raw data alone is not enough.
- Beware of extrapolating into the future. Similar to the above,
just because wavelet coders produce better numbers than fractal
coders do now, that doesn't mean they will tomorrow. Each brand
of technology is in a different stage of development. Some are
young with much potential still to be exploited; others are
mature with most optimizations employed. Estimating how close a
given technology is to being fully maxed-out requires detailed
knowledge of the field.
- Beware of JPEG. By virtue of being an international de jure
standard, it will be hard to displace. Anything else must be much
better, not just somewhat better, to stand a chance of widespread
popularity.
- Listen to your eyes. (How's that for a mixed metaphor!) Distortion
numbers are presented for mean absolute error, root mean square
error, and peak signal to noise ratio. These are well known, easily
computed, and central to much of mathematical analysis, but relate
only loosely to human perception. One of the hopes of the BragZone
is that it facilitates the derivation of superior distortion measures.
- There are no timing results. The speed of compression and
decompression are also critical qualities. As are memory
requirements. Really, the graphs should be plotted as compression
vs. quality vs. speed vs. space. But that's tough.
- "Best" is a complicated thing. Ranking coders for a particular
application involves a delicate weighting of many factors. Top
among these are compression vs. quality, compression speed,
decompression speed, and memory requirements. Also, the types and
sizes of images that can be handled. Then there are computer-related
issues such as: platform availability, portability, ease of use, ease
of integration, executable and/or library and/or DLL form, source code
availability, and source code language. Then there are business-related
issues such as: program origin (commercial vs. non-commercial), maturity,
support, royalties, patents, and so on. Reaching a "best buy" decision
after sorting through all the many considerations is what consultants are for.
Go To
[Fractal Page]
[BragZone Page]