Re: 20 Years of JPEG Celebrated With Software Launch

Guido Vollbeding wrote:

And you continuously claim so, and failed to deliver any proof for it.
If this is the central point of your argumentation - and if it is - then
please provide facts that backup your claim.


Thomas, again: This is not a point to argue, and I cannot backup my claim
on basis of what you would accept.
I cannot give you my eyes for vision - you can only use your own eyes.
And that's what I recommend: Just use your own eyes.

So do I, Guido, so do I.

But instead you are still searching for and demanding better measurements.
It doesn't work that way.

It works exactly this way. That's the way how science works: By finding
objective grounds for claims. Otherwise, we can of course agree that "JPEG-1 is better for you than for me", but what can we derive from this insight? Nothing.

You must *first* see for yourself, and *then* might find a good measurement
someday which matches your visual experience.

So do I. I stated more than once what I'm able to see. Now, aparently, we agree to disagree, so we need some objective ground, otherwise there's no arguing.

But without visual experience all your measurements will be useless.

Except that I personally cannot make *objective* measurements. I can make *subjective* ones. Very nice, but only helps myself. And as my subjective results are different from yours, no conclusions whatsoever can be drawn. Thus, no knowledge gained, no lesson learned, and no clue whatsoever what to do better the next time.

I have a visual experience which is more worth than any of your measurements.

Why do you continuously disallow me to make my own subjective tests? My visual experience is just different. Don't you think I've never looked at the images? Why are Guido-tests the only acceptable tests?

Because I had to look at millions of images over many years. And I'm looking
nature, so I can well recognize defects and differences.

You're trained to JPEG-1. Of course. So that's a lousy test. You need unexperienced people, or customers from the street, if you want to say so, and ask them about what they see and don't see. Otherwise, you will look for artefacts you're not familiar with, and I look for artefacts I'm not familiar with, and in the end we cannot conclude anything.

Point is, I cannot run these tests, I don't have the means to do that. I can neither pay the equipment, nor the people. That's exactly the regime of full-reference objective computer based tests - and a research field. You seem to deny the usefulness of these tests as only the Guido-metric is the real metric. For you - of course it is. For anyone else - likely not. You cannot generalize from your own visual experience, that's the point.

Does not well for small images, does not well for high-contrast images.
Does well for large images, does well for flat images. I also explicitly
stated this in the ISO/ITU meeting, and I explicitly made measurements
to see this.

("Well/bad" heree as whether there's an advantage to JPEG-1.)


So that's why we make efforts to improve JPEG-1 on the "not so well" cases
(see "Sudoku" extension proposal to address large images, for example).
Why shouldn't we do this?

I've never said you shouldn't do this. Of course you should! Of course WE should. Of course the ITU and ISO should in common. But for that we first need to agree what is "good" and what is "bad". And for these tests, "Guido" is a bad visual standard. Not because of you specifically, but because it is a norm that is based on one single observer.

We have a well established technology which does
well in many cases, and now we are going to improve it for other cases.
What's wrong with this way?

Nothing's wrong with doing so, except that you're so focussed on one single technology. That's why I say that you yourself are the largest obstacle in finding improvements - you wouldn't accept anything but DCT. And that's a very bad starting position for making progress.

For example, why was ITU-T.851 necessary? There's no new idea in this standard whatsoever.

Why should we throw away our established
substance and switch over to something completely different only because
some strange people present nice PSNR tables and rate-distortion graphs?

Quite the contrary, I don't say that you should throw things away. Keep JPG what it is! It does a good job! What I'm saying is that if you want to improve, you need to widen your horizon. Of course any new standard must measure against JPEG. It must also measure against JPEG2000. And measure definitely also means "beyond PSNR". But I *insist* on "measure" and not a "Guido metric". Of course, people providing the metric must make sure that this has a good correlation to visual impact. And of course just measuring against PSNR is an error. PSNR, however, can *also* tell a story, namely whether the overall codec design is sane, as there are a couple of interesting theoretical results on it.

Again, the point is not in having nice-looking PSNR tables and
rate-distortion graphs, the point is in having nice-looking images.
When will you realize this?

Since long ago, Guido, since long ago. There's nothing to argue about that.

If you have images where you find that JPEG2000 is worse, it would then
be interesting to know which codec this is, and in which mode it is
driven. Unfortunately, due to a lot of "marketing hype", people push
PSNR too much and thus use the codec in PSNR optimal mode. This is of
course wrong. And again, people use the jasper code, which is also not
an optimal choice. Actually, it's IMHO the worst choice, but then others
might have other opinions.


The jasper code was just the first code which was freely available.
I see you constantly complaining about bad J2K codes people use.
But why don't you care to provide something better instead of complaining?

I do care. The committee cares. For example, the UCL has an initiative about an OpenJPEG codec (google for it). I haven't looked at it, but these are people that know how to do their job. You can also check with the JJ. Otherwise, several notably better codecs exist on the market. You can also get a test version if you need to - thus claiming being unable to find a good one just means that you're unwilling to look, I afraid.

Your complaints are simply useless as long as you can't recommend
something better.

A recommendation. Ok, fine, here we go, my personal hit-list (from best to worst.)

a) The Pegasus codec. (-: Well, that's mine of course. No surprise. Note the (-:.
b) The Kakadu codec. I think you might get an academic licence as much as you could get a test version of "my" code.
c) The JJ2000
d) The Jasper

where c) and d) definitely in that order. I haven't tested against OpenJPEG myself, so I can't make a suggestion where to put it on this list. Probably between b) and c) as I'm aware of a couple of minor JJ bugs.

And what is the purpose to have a standard which looks good on paper
but is useless in practice? It doesn't work that way - there are plenty
of useless standards on paper, but what counts is whether and how they
are applied in practice.

Huh, since when doesn't it work? Of course it works. It's also sold, but in different areas.

And that is the point of the Independent JPEG Group:
We choose a reasonable standard, and make it available for application.
If you don't have something like that for your standard, then it is simply
obsolete.

Of course we do have something like that. You can buy JPEG2000 from a couple of people. University of New South Wales, Kakadu (IIRC). Pegasus Imaging, Aware, LuraTech. You can get test versions for free from most of these people.

And I have told you why I am not going to support your standard - I simply
don't consider it reasonable.

The problem is that this is pretty much prejustice. You failed to provide *objective* reasons. Instead, you consider it cool to run personal attacks against anyone who doesn't share your p.o.v. - it doesn't work that way. My view on all the claims you've made here is that you've now maneuvered yourself successfully into a position of no return. Stop it, for yourself.

And apparently nobody else appeared so far to do something for your standard
similar to what the IJG does for the JPEG standard. That's your problem.
Please think about it, there are reasons.

There are reasons. To some degree that there's no widely available free codec, right. To some degree, too, that the available FREE ones are - well - not really what I like them to be. To the major degree, however, that the problem JPEG-1 attacks doesn't exist anymore.

Consider we had a standard - just consider - that would compress the images to half the size we archive with JPEG-1/JPEG2000 today. (No, that's not Microsoft - that's just marketing.) Would this be successful? My personal opinion: No. There's no point in doing so, because there's no problem in having a 100MB image. Go to the next MediaMarkt, Cicruit City or Radio Shack or whatever, buy the USB stick with 4Gig, problem solved. (For M$, the situation is of course a bit different, they can
force people to use whatever they want people to use, but that's then again a different story.)

Problems lie elsewhere today.

Should we stop doing image compression? Maybe. But possibly, we should understand where the problems lie, and should design compression schemes such that they solve the problems the people have. JPEG2000 part 9 is a good example where this principle applies. "Just" compress my images, well, JPEG-1 does it, no big deal.

And instead of supporting J2K, I have found my own ideas how to improve the
given technology for more applications. That seems reasonable to me, and
nobody else seems to do it for me, so that's why I'm going to put that in
a standards extension with ITU.

Well, except that the market niche you're targetting at is probably occupied already. Which "new problems" are you able to solve you cannot solve with existing technology today? As Jack put it: "This boad sailed away long ago." It's really that. ITU-T.851, for example, was pretty much pointless after all.(Any new solutions for new problems? No.)

So maybe you do have new ideas, why not. As said, I'm not the one who will stop anyone, I can only learn.

I think you ISO JPEG and JPEG-2000 folks have just a different approach
of how things should work, but apparently things do not work that way.

Aren't they? The point being is that you seem to have no clue how things do work here at ISO, so why do you imply this? It's an open committee, ideas welcome. Just not warmed-up old ideas. Make something cool, new, prove that, and you're welcome.

I have made some experiences of how things work - I see the reality,
I see the success, and I see how to continue that success story...

Shrug. I see the success of JPEG-1, but I don't see how drilling that idea up to infinity should have success. But then, I don't stop it, go along and see yourself. What I'm seeing is no real improvement on this road. It requires smarter ideas, and especially, it requires finding the problems people really have.

Here's an idea for you, and something ITU and ISO should probably care about: Standardize something like Open-DNG. An extensible, well-accepted, open, raw-image file format. Wouldn't that be something cool to have, based on the official grounds of ISO and ITU?

So long,
Thomas

.

Relevant Pages

  • Re: 20 Years of JPEG Celebrated With Software Launch
    ... Because I had to look at millions of images over many years. ... And that is the point of the Independent JPEG Group: ... We choose a reasonable standard, and make it available for application. ... Please think about it, there are reasons. ...
    (comp.compression)
  • Re: A really good article on software usability
    ... Thus not allowing images ... not feature). ... sheets with multiple javascript pages loaded through definitions. ... standard somewhat less than a standard or we could blame the standard ...
    (Fedora)
  • Re: [SI] 2008.08 is up
    ... Or why not just critique the one's you particularly like, ... Why, for exhibiting images on the web, *should* there be any aspect ratio ... The square aspect has long been a standard. ...
    (rec.photo.equipment.35mm)
  • Re: XDepth Raw: a jpeg-compatible Raw compressed format for digital cameras
    ... Uncompressed HDR images are generally 96bpp images. ... but I'm not trying to advertise 8bpp for HDR. ... is an ISO standard, you can make sure that you get your images back even ... The same cannot be said with your format. ...
    (comp.compression)
  • Re: D STAR - couldnt have put it better myself.
    ... What shouldn't become the only standard? ... similar open sourced codec is going to become the defacto standard. ... So tinker fiddle and improve to develop an alternative, otherwise leave those who are happy with DSTAR be. ... if those wanting an alternative do no more than say they don't ...
    (uk.radio.amateur)