Re: BTPC Extension - PyramidWorkshop

niels.froehling@xxxxxxxx wrote:
Rough estimates are that the PW beats best-of-PNG+JPEG2000 in 50% on
all cases (... that I checked: http://paradice-insight.us/corpi/) in
lossless mode. Sometimes the win is very spectacular 50% of a PNG for
example (fractal images with chaotic structures win a lot).

Well, according to the posted measurements above this happens in lossy
mode only for non-natural, computer generated images (logos, etc.) with
lots of flat areas. For the lossless mode, I'm not really surprised, but
then a codec tuned for lossless should enter the comparison, namely
JPEG-LS. It is based on a surprisingly simple adaptive predictor.


I hope you're also satisfied with the CALIC+ari results. I may repeat
that with
JPEG-LS+ari if you insist. :-)

Well, I think you definitely should test against JPEG-LS here. Because it is a well-acknowledged algorithm and you will sooner or later hear questions about your perfomance against JPEG-LS anyhow.

Also I hope the results show you that the PW does not have these
properties (being tuned for flat), I would also 'claim' that PW is the
most _stable_ algorithm in the results shown (it never completely
sucks, like CLC in one or two cases, or JP2 in one or two cases).
I hope I can present some lossy results in the next time.

Would be useful, definitely.

Possibly not, but then again I haven't seen these measurements either.
BTW, why do you want to use subsampling in a pyramid-oriented image
compression setup? Furthermore, if that should be a fair comparison, all
codecs should be driven by the same subsampling factors.


Because PW takes care of inter-plane relations, if it detects shapes
in one plane it applies that knowledge to the others, that means with
4:2:2 subsampling and existing inter-planar shapes, the loss is
extremly small. I play the strength of some aspects of PW against the
other algorithms, I assume that JPEG-2000 do not catch inter-planar
properties.

Not really. Inter-plane correlations are only captured, if at all, in the color transformation. Afterwards, planes are coded independently. This step was made to make the algorithm more "parallelizable" (is that a word?), and possibly to fit into the traditional scheme of image compression, too.

I think playing strength out is not 'unethic', JPEG-IJG took
subjective advantage of the standard setup YCbCr 4:2:2, which probably
is the strength of JPEG-IJG.

Clearly not 'unethic'. (-; Question is what does it buy?

I know you have to distinguish
between low bit- and high bit-rate, I mean high bit-rate here (> 1bpp).


For other against JPEG2000, look at high bit rates, I think it's very
hard to 'beat' JPEG-2000 on low-bitrate. Maybe with some sort of
adaptive neural vector-quantizer, low bit-rate is the domain of
vector-quantizers.

Yes. The trellis-quantizer in JPEG2000-2 causes still quite an improvement, but that's also mainly in the low-bitrate area. JPEG2000-1
uses only scalar quantizers (quite unfortunately, due to patent restrictions of the trellis/Viterby algorithm).


Automated test programs should be the least problem, I've something
handy. Just were can I get something that runs on Linux? (-;


That, my honorable listener, is the reason why I made it public, I
still may have the time for bug-fixing in the core (who knows it that
good), but we (hey it's public know, it's now yours also :) may have
the time to make it usable everywhere.

Well, and "testable". Which is a lot more than many people do here (namely, post claims and do not provide software for testing.) Thanks! I cannot make promises as *when* to look into it as I'm rather booked out, but I will definitely look into it.

So long,
Thomas
.

Relevant Pages

  • Re: faster rendering of eps pictures in YAP?
    ... Images on the hard drive take more time to load than images ... by the cpu to implement algorithms, ... So using a more efficient algorithm saves time. ... Lower resolution and smaller images will take less time, but then quality will be sacrificed. ...
    (comp.text.tex)
  • Re: Multi-sampling and "2400x4800 dpi" scanners
    ... than the same picture taken directly at 2400x2400 dpi from the scanner. ... Sounds like some pretty hefty algorithm to avoid ... though there are some images that work better using that. ...
    (comp.periphs.scanners)
  • Re: New lossless image compression algorithm
    ... Well the algorithm should be quite stable, ... "Lossless Compression and Interpolation for High Quality Still Images ... whereas LOCO allready is in cameras ...
    (comp.compression)
  • Re: BTPC Extension - PyramidWorkshop
    ... example (fractal images with chaotic structures win a lot). ... For the lossless mode, I'm not really surprised, but ... I think playing strength out is not 'unethic', ... Well I only made an example that PW does not suck in low bit-rate as ...
    (comp.compression)
  • Re: finding dupe files
    ... Finding duplicate files is easy - there are hashing algorithms that ... term, and computers are horrible at dealing with fuzzy requirements, ... If you're talking about an algorithm that would detect and group ... still images, though it might be an interesting experiment. ...
    (alt.photography)