Slashdot Mirror
Yet Another New Image Format
An anonymous reader sent us a link to a PC World story
that talks about a new
wavelet based image
compression algorithm that (you guessed) produces smaller
images at a higher quality. As usual, I'll believe it when
IE and NS integrate support for it and webmasters use it. I
bet we're still using gifs and jpegs years from now. What a crock.
In the interest of self-enlightenment, I will ask
this question: what is "alpha transparency?"
--Anonymous Coward
It may be better to remain silent and seem a fool than to post and remove all doubt, but I gotta know. What, exactly, is an ROC curve?
Weblogging Considered Harmful:
Wavelet based techniques are much better - less artifacts (no blockiness and other crap), smaller. ;(
But - most of algorythms are proprietory
BTW. The example in the post above - with a faint line - is exactly where DWT based techniques shine - they preserve singularity type structure in the data. Look up publications on wavelet denoising.
Good start page
<^>_<(ô ô)>_<^>
... besides web. DCT (eg JPEG) is good for natural images generally but there has been some sound research done on wavelets, especially for specific image classes such as various medical imagery. Just because commercial web browsers don't support an image format doesn't mean it's useless.
My 2c worth.
The company I worked for was using wavelets from Summus (we're both in SC). I talked my superiors into dropping them when we went to port from Windows to Macintosh cuz I did'nt feel like getting their Mac SDK. It was very proprietory and very slow. I got my company to switch to JPEGS cuz JPEGS are 10 times faster (maybe more I did'nt profile it) and for equivalent file size the only noticeable difference is that text on JPEGS is a little blurrier (the images are clearer though). I have'nt used the toolkit for over a year, so maybe they've updated, but I'm waiting for the new JPEG format to be established cuz Wavelets bite the big one.
I may be using the term incorrectly, but I take it to mean having different levels of transparency. The GIF format only has simple transparency: either you can see through the GIF completely or not at all for a given pixel. The PNG format has 256 different levels, ranging from fully transparent to fully opaque. The intermediate levels would be nice for blending an image with the background. For example, the background of a web page would show partially through a shadow in another image. Most web design books show examples of a graphic with a drop shadow designed for one color that looks awful on a background of another color. Alpha transparency would reduce that problem dramatically. Unfortunately, I believe that all browsers which currently support PNG graphics convert the alpha values to simple transparency, and rarely very well.
Weblogging Considered Harmful:
Wavelets are, from a mathematician's viewpoint
a generalization of Fourier formalism. They
represent an area of mathematics and are hardly
related to any specific algorithm.
Many wavelet algorithms are published and they aren't just useful for compression (I remember
writing an edge detection program based on one
of them). IIRC, many names among the founding
fa(mo)thers in the discipline sounded French
(e.g. Daubechie) and it is interesting to know
what important developments were INTERNATIONALLY
patented.
Your statement about wavelet compression's effect on ROC curves is not supported by my observations. Can you provide a reference?
Thanks,
I.P.
there's plugins for NS and IE to read wavelet image for years! even a good friend of mine in university in 1995 make in C a simple compression/decompression program for sound using wavelet... wavelet can be used to compress sound very good too.
--
"Science will win because it works." - Stephen Hawking
They've either been smoking something, or failed basic biology, cause this ain't no butterfly.
This isn't the first such fractal-based ...)
image compression encoding format that has
been promoted - there have been several others
over the past year or more. Alas, all of them
have been proprietary. There may be a niche
market for such a product but they haven't
seen widespread use for a good reason.
(And all of them have been too closely tied
to use on windoze
BeOS/netpositive uses PNG transparencies.
The OS supports the alpha channel so implementing it in your software isn't so hard.
-josh
It looks like AWeb supports it just fine.
Well, at least, when I go to
http://www.cdrom.com/pub/png/pngsuite.html
everything looks like it's supposed to, including the alphas.
is what it won't work on. There are many types of images where the pixel-to-pixel statistics decorrelate so rapidly that the images are essentially random-like (that is, they have high entropy with respect to the feature of interest). Compressing these types of images with methods like wavelets causes major losses in the ability to retain the faint features that drive the detection statistics. For the technical minded, the ROC (Receiver Operating Characteristic) curve is damaged. That is, the probability of correctly identifying (or even observing) features declines.
So, prepare to be disappointed.
Suppose you have an RGB colored object which is partially transparent (eg, a balloon or colored
glass) and you want to compress an image of the object and still have it's partial transparency
represented in the compressed image. You need another component, thus RGBA (A is alpha).
Alpha is also used to make the background show through non-rectangular objects and to anti-alias
object edges.
The reason it is called alpha is that it turns out it's useful to model transparency as a blend
factor X*(A) + Y*(1-A) so that if you put multiple transparent things overlapping it looks like
real life.
This is in contrast to a transparency bit to indicate if the pixel is transparent or opaque.
Often this is called 1-bit alpha.
You can simulate alpha transparency by dithering the 1-bit alpha, but it just isn't the same.
Unfortunately, you forgot to include the URL for the Analog Devices site, specifically related to the adv6XX chips.
Here 'tis:
http://www.analog.com/techsupt/software/lcm.zip
...Open Source isn't the only answer -- but it's almost always a better value than the alternatives...
PNG v1.0 was declared 'done' May 30th 1996. It achieved official recognition status from the W3C in August of 96. So three years isn't all that inaccurate.
If you want to experiment with completely GPL:d image compression algorithm, you can download my GWIC (GNU Wavelet Image Codec) image compression algorithm. It is certainly alpha quality and the compression performance is not of highest state of art (but neither is WI :) ). In fact the compression performance should be somewhat comparable to WI.
I have not done any development for GWIC lately, because of the lack of interest, but have already almost ready to use new version of it, if someone is interested in integrating it to some other program. GiMP anyone ?
It is quite easy to add progressivity (in fact the format already supports it, but I have not implemented probressive decompressor), regional focusing (in fact I have already implemented that some years ago in another compression system) or alpha channels. Also I have implemented distortion limiting to the compression (not integrated into GWIC yet), which allows the user to specify the exact quality of the image, not only the quantization or target image size
Vaadin - the best open source framework for building web applications in Java - no plug
OmniWeb3 (a NEXTSTEP, MacOS X Server and Yellow Box/NT browser) supports it. I just looked at the PNG test page and we apparently fail the "Images with non-square pixels and/or pixels with physical dimensions" test. Guess I'll go fix that :)
:)
http://www.omnigroup.com/Software
(and yes, I work for them -- heck, I own them
kaboom!!! ===(my brain)
There seem to be many wavelet based formats coming out. Check out www.lizardtech.com and www.luratech.com for 2 more proprietary wavelet based formats.
Also PNG is not a lossy scheme and the authors said they will never put a lossy scheme into it. This is good, because we want to see ".png" and know it is lossless, and see ".jpeg" (or ".wl") and know it is lossy. Do-everything things like tiff are doomed because we want to know what is going on...
Have you tried to surf todays web using Netscape or IE 2? Half the web sites do not work, and a lot of the ones that do look terrible. People will upgrade their browsers if the web becomes unworkable with their current one.
How many people do you see still using Mosaic because they are affraid Netscape is buggy?
A researcher at USC's SIPI (signal and image processing institute) proved that fractals are just a subset of wavelets. So its not really possible to say that fractals can do things that wavelets can't.
-BK
...they are closer to fractals - zooming property is linked to self-similarity. That's what makes them good for edge detection and compression.
<^>_<(ô ô)>_<^>
I know someone on the JPEG 2000 committe and it
will all be wavelet based compression.
It's state of the art in compression.
Notice this is very Windows-centric (uses an ActiveX plug-in) and not open.
Folks, I don't know about you, but I am very happy with JPEG. The format is well documented and supported by many open (and closed) source applications.
I can't see how any company can come up with a replacement for JPEG unless it is released entirely to the community, free of royalties or proprietary licensing.
Here you can find a good wavelet tutorial, "The engineer's ultimate guide to wavelet analysis". With it, in 2 hours I understood the essence of wavelets ; contrary to many wavelet books, done by mathematicians who spent their time showing how smart they are by detailing the demonstration of every single property of wavelets that uses sophisticated mathematics, instead of even parenthetically starting to suggest how wavelets could be used in real world ; and who use an excess of formalism instead of detailing the intuitive notions behind wavelets.
PNG is just -starting- to creep into mainstream use. And I believe its at least a year old now.
PNG has been out for YEARS. The author of the old DOS image viewer QPEG created it.... 5... 6 years ago? It's just you haven't seen it because it was made to kill GIF, something that just wasn't going to happen in the large BBS scene and the 40 billion GIF viewers that were out there.
Only when compu$erve decided that it was going to be a cock about it's proprietary image formats (at least AOL has been stringent from the beginning) did people start considering PNG.
-Erik-
AT&T's research arm produced a wavelet algorithm for encoding scanned text and line art. JPEG introduces horrible artifacts when it compresses line art or scanned text. I have been using this solution for a genealogical group with good success. They've even got a linux version of the compressor and decompressor. Check it out:
http://www.djvu.att.com
or email me, warnereng@earthlink.net
Why ?
Can you tell us why the *HELL* it's useless if it's not open ?!!
I'd like to know what kind of knucklehead tries ;)
to compress an image file of random static. I
don't know of many algorithms that would handle
that well...
CIC3 at Los Alamos Labs has been doing this wavelet stuff for a long time. They have a demo online at:
http://www.c3.lanl.gov/~bradle y/mrsid/washington_dc/It is a lossy scheme (almost obviously, since I haven't heard of any commonly used integer transforms other than Infinop and those using the Sweldens' Lifting Scheme).
The issue of color isn't a big deal-- if the tranforms can be applied to gray scale then color is just a YIQ(possibly) transform away.
Ermm..as for alpha transparency-- I don't know if that ties in directly with the compression scheme since it _could_ be implemented as a separate mask. But considering the way the transforms work (look at JPEG)... it would be a mess handling the edge effects involved in good alpha'.
Is the standard open?
From what I saw on the web page, it didn't seem like I could find out how the compression worked or how to write my own viewer, but maybe I'm wrong.
Thank you for the lesson. The argument is so clear that I feel a little disappointed in myself for not being able to think of it on my own. I guess that's the way it is with a lot of things.
The "cue the foo posts in 3, 2, 1..." posts will commence with no subsequent foo posts in 3, 2, 1...
>A .tar.gz of BMPs would probably be smaller than >a lot of single GIFs
Two potential shortcomings with this. One, over how many bytes does gzip check to see duplicate patterns? I remember 'ol pkzip for dos, unless otherwise specified, only checked 64k chunks in the name of compression speed.
Second, tar.gz is made to compress binary files. If you can develop a lossy compression algorithm, one that is right-brain recognizable, then you do not have to worry about being as accurate with the data as gzip would be. This is why gzded BMP's don't compare.
it's think that it funny that the people screaming that wavelets are a proprietary non-open source(tm) piece of shit are also the same people creaming in their pants over mp3's, another non-open source(tm) proprietary format. before someone calls me a loony, doesn't the mp3 format have a few patented aspect (ie, proprietory)? didn't the patent holders sue/beat/disembowel the people whom released a gpl mp3 player. if so, then no open-source version exists. funny, downloading a gpl'd program from a warez site.
i've used wavelets. i'd played with the "voodoo dejavu" technology out of at&t. they are cool. as to this "you need a plug in for them", yes, for now. also, the various formats are incompatable. i remember jpeg also going through the same thing. cshow and gws each couldn't reach some jpeg files that the other could read. then, the standards committe put down their foot and we finally had one format. wavelets will be the same way. maybe jpeg will incorporate wavelets in the v2000 spec?
also, wavelets are much better at video than mpeg/motion jpeg/avi files. this is one of the things we're dinking with right now. you can re-wind them, whereas you cannot rewind the *peg's! avi can possibly rewound, but the file is huge (and proprietary). wavelets are smaller than all the other.
is our dinking around using a proprietary format? yes. once some committee standardizes on something, we'll change our format to that. mpeg-4/5 would be cool if it had this.
gzipped data can be decompressed without waiting for the whole thing to arrive, and tarballs can be read sequentially (unlike the moronic pkzip format that puts the header at the end, rendering incomplete transfers useless). So, it sounds like .tar.gz image packs are perfect for the job... Perhaps someone should add an extension to mozilla, perhaps in the form of a supported meta tag to reference a tarball containing all of the images used by an html document...
- RF (dfelker@cnu.edu)
Posted by Mr. Assembly:
Stop askng if is open - It's probably patented. My guess is that it could lead to the hi-jacking of a format lke what happenned to GIF a few years back.
People probably would not want to hassle with another plugin anyway.
Most, if not all, of the Amiga WEB browsers have supported PNG for quite some time now. It's about time people switched.
However, lets see what WI is like - maybe we'll be slating PNG next year...
From compresion 101: This is always true by what is known as the counting argument.
Suppose we have a 2:1 compressor that takes 1K images down to 512byte images. There are a
total of 256^1024 possible images, and we only have 256^512 representable (assuming our compressor
is perfectly efficient). That leaves about 256^1024 - 256^512 ~ 256^1024 possible images not representable
which is almost the same amount you started with.
This means you pretty much have all the possible images to form your interesting class of images that kill your compression algorithm.
Compression is easy, you just have to know which images you don't care about (which unfortunatly is
the hard part).
What, exactly, is an ROC curve?
You've heard of the "Graduate Student Algorithm" for compression, right? Well, somebody recently came up with a slightly simpler implementation that doesn't require grad students; almost anybody can take part. So they're looking into hiring the entire population of the [Peoples'] Republic of China to work on these things. They've done some research, and they've found that certain types of images interest the workers more, and so they get compressed faster and better. These preferences have been charted. This chart is called the "ROC Curve", where pictures of the marketing secretary's new kitten in a cute little basket are like way down on the bottom, and pictures of a rather more prurient nature tend to clump at the top. Keeping in mind the relative commercial significance of these respective things (and my own tastes), I think we can say with confidence that this phenomenon is no great hardship.
The primary concern, of course, is that the algorithm should perform equally well at all points on the ROC Curve. This is because if the top-end images get mangled, interest will wane, which will further degrade image and compression quality. The end result of this distortion of the ROC Curve will be a vicious circle of lowered expectations and disappointment for all concerned.
It may be better to remain silent and seem a fool than to post and remove all doubt,
. . . and then again, it may not
This looks very cool, especially for images and video. What would it take to use the same mathematic priciples and to have an OSS codec?
Any OSS codec would kill a proprietery standard due to it soon being ported to greater platforms than just Netscape / IE under Wintel.
Codecs are commodaties, and the sooner the world realises this the better.
Ice Tiger
"Because we are not employing at entry level, offshoring will kill our industry stone dead."
It's taken 30 years, but JPEG and PNG(zng, jzng, jng, mng, all the pnglets) are all that matters any more. There isn't anything this new format will do that isn't already done by jpeg or png or won't be done by jpeg2000 or jng (or a newer png)
...is one of those r33t-a$$ w3b d00dz who writes: "Go in your controll panle and change yor monitor too 1024x768 too view this page and b sure you pick 16.7 milloin colorz if u dont u can suck my very extreemly big c0ck cuz you must bee stoopid or sumthing!!!" on his page(s)...
*sigh*, lamers...
- RF (dfelker@cnu.edu)
from the person without a brain---
I did dude, i know what it entails and how each works and its differences... my point is, DUHHH, any new fangled format has 0.00001% chance of being successfull, im not insultinga nyone, just provoking a statemen.
Do we all have to be polite and good little nerds, or are we allowed, YELL OuT and go wild sometimes....
There you go...
Can you provide a good pointer as to where I can
learn more about ROC curves?
AT&T also has an image format that compresses images smaller than jpegs. It produces images about 1/4 the size of jpegs. Its proprietary and looks like they're trying to make money off the tools to create them. Although they do have a plugin though for Navigator under Linux. The link is here
http://www.djvu.att.com
So the neet fact that the image is smaller is basically crushed by the time/space/bandwidth it takes to install proprietary extensions to decode said "small" image. Foo.
At some things at any rate... Can you say somewhat lossy but still usable 300:1 compression for real-world picture? (sorry if that's backwards)...
Apparantly there are a few formats already in use, but all are proprietary.. Would an Open format be possible despite the software patents that have surely been taken out by now?
Take care folks
-MJayM
a lot of the usernames are the same. same writing style for some of the ACs. hell, mr. taco rips on wavelets and drools over mp3s.
Just one small problem. Compression requires a Beowulf cluster running overnight ...
lake effect weblog
{Network engineer in Chicago--looking for work!}
Well, It all depends on the filter pair used to do the wavelet decomposition. Wavelet compression's ultimate goal is to remove coefficients that represent details. If a filter pair is chosen that captures some details in the low-pass portion (approximation portion), the result will look very reasonable for the output size, however, the filters that are less mathematically intensive (faster and easier to implement) generally destroy detail and look like crap.
Wavelet de-noising is pretty neat. But there's
still a thresholding issue. You throw away
all sufficiently small wavelet coefficients.
If your details involve small variations
you'll lose them along with the noise. But it
does beat the hell out of assuming anything
high frequency is probably noise and should
be filtered out. The edges end up looking pretty
good with wavelets.
I was just reading a paper on generalized
cross validation for selecting the threshold for
wavelet de-noising. I was impressed.
Yup, not terribly new. I was playing around with one back in the 96/97 school term; though at that stage it was still pretty crummy quality.
Here's an idea that I haven't heard of yet to help compress still images.
:), I haven't had any experience with this yet) they were compressed to find patterns against each other as well? This would be perfect for webpages which show more than one image at a time anyway.
:) )
What if images came in packs, and after being compressed (using an algorithm
I understand mpeg does something like this, but that adds the fourth dimension (time, there is no third dimension here
This type of stuff is not for you guys to use tommorrow on some website project...it's to ease the struggle of those of us that deal with HUGE images that look like crap after a run through JPEG math.
Photoshop does some incredible stuff -- but if you do pre-press or catalog work, these images can be REALLY big.
I just don't give a rat's rectum about technology for web graphics. Where I work, we still have to design for Mosaic and Netscape 2.x. This stuff is cool enough to get someone burning 1,000,000's of images that are print quality on a CD -- that I can then browse using a non-proprietary web browser to see.
Grow up boys -- look beyond technology that leaves you typing with one hand.
FAST Image Transfer, it's made by the same guys that gave us FAST FTP search and FAST MP3 search!
i d=44
Take a look: http://web.fast.no/product/imagetransfer/det.asp?
What the...? I think I saw a penguin...
Mother Wavelet comes to me, singing words of wisdom: "jay pee gee, jay pee geeee . .
:)
Greater compression on images only addresses bandwidth issues and completely neglects latency.
.JAR, then pages would appear to download more smoothly, especially ones with lots of small images.
:-D
We already have JPEG, and and if anything replaces GIF it will be PNG (the sooner the better!). PNG handles "solid" colors well, like GIF, but its an open standard like JPEG is, and supports 8 and 16-bit alpha... rather than the 1-bit alpha "on" and "off" found in GIF 89.
HTTP servers typically start a transfer off relatively slowly, allocating more bandwidth as a file progresses. So unless your image files are really large the server is never reaching potential because it finishes the xfer before increasing the bandwidth. IOW, servers do better with 1 100k file than with 10 10k files.
If the images could be bundled into a container format, like how some Java applets use
Another big waste is "localized" websites that are not at all local. It would help everyone if sites like "Yahoo Boston" were ACTUALLY located *IN* Boston so you weren't dragging a page across the country. It's wasteful. Maintaining a remote webserver is very easy to do if you use UNIX (Yahoo uses BSD..).
Of course, Microsoft's Sidewalk sites can't be remotely maintained because they run on NT [network farms...]. LOL... I wonder what MS' IT budget must be, aside from the fact that they are exempt from hundred thousand dollar NT server licenses...
Nothing will do more to help the web than local caching.
He specifically said "either have incredible loss or a negative compression ratio" (that is, the file gets bigger), which is the case with LZW on purely random data (the best you can hope for is to almost break even).
---
"'Is not a quine' is not a quine" is a quine.
Quine "quine?
AT&T's DjVu though not "open" is still much better than this stuff (Atleast for scanned content). Why?
1) There is a Linux Netscape Plug-in for viewing (not Windows only).
2) There is a Linux (and other Unix) compressor that is free for content that meets the license requirements.
3) The compressor/decompressor works with pnmtools.
4) It can turn a 9MB gif into 90k in djvu format retaining the high quality.
See http://www.djvu.att.com.
LZW (which I think is the heart of the UNIX compress command -- pls don't flame me if I'm wrong) would give lossless compression, if it compressed. But, the "compressed" file is as large as the input file in this case.
All this is sort of an informal way of saying that the degrees of freedom needed by the compression algorithms is equal to the number of pixels in the image. Hence, no compression.
Sorry, I couldn't resist.
We've occassionally used the making of foie gras as an example for this topic. You stuff alot in and hope that the little bit that comes out is edible.
This not a case of lossless compression, however. The result is vastly different from the input. But, you're sure willing to live with it!
As long as it's supported by Microsoft software, people will use it. Microsoft will love the idea, because it's another way to trap people into using their non-standard "standards". Excel and Word do it very well already. About time they get ahold of a proprietary image format. Wait for this company, or this algorithm, to be bought by MS.
Ermm... just going by your average wavelet transform (perhaps using Lifting Scheme) they are linear time transforms (that's the neat-o factor about it).
PNG is a lossless format, and as such will NEVER
:-) it wont go
;-).
compress to the same level as JPEG.
JPEG on the other hand, is lossy, and so will
NEVER retain the image quality that the PNG does.
The format mentioned is a pretender to JPEG's
crown of being the lossy compression of choice
on the web. since JPEG is supported in nearly
all browsers (I think that lynx still haven't got
their jpeg support working yet
away (like GIF).
PNG is an attempt to rid us of the proprietary
compression algorithm headache experienced by
GIF users, and NOT a replacement for (or
alternative to) JPEG (and thus NOT a suitable
alternative to this WI). That said, there are
other wavelet based compression ideas around, and
rumours that the next JPEG standard will be
wavelet (and hence WI is Web-Irrelevant
Well, the FBI chose wavelets for their fingerprint compression standard because they beat JPEG rather well at low bit rates (see http://www.c3.lanl.gov/~brislawn/FBI/FBI.html). Wavelets work well for "natural" images, but they don't typically work well on "synthetic" images, like text, cartoonish images, etc. Maybe the one you were thinking of was trying to use them on the latter?
No.
Summus owns a stack of patents on wavelet codecs, and they're royal bastards about "their IP". In fact, this is one of the most patented areas of mathematics. Even though you can't patent mathematical formulae under US law...
Wavelets, especially adaptive wavelet coding, is vastly superiour to DCT (the basis for JPEG and MPEG), especially if you do video codecs using 3D wavelet transforms. Why haven't they taken over? Because it's a patent minefield.
I saw this stuff three years ago - it looks nice, and it really does work, but it's not remotely open. The JPEG2000 standard will likely be no more open and free than GIF or MP3, with their associated patents.
ATT's Dejavu seem's to be great, excelent compression ratio, quality, etc.... Why it's in the shadows?
The viewer can be found of off Summus's site:
d .html
To download:
http://www.summus.com/products/download/downloa
Click on ActiveX or Netscape to take a look at there image gallery.
Amoeba
Webmonkey has a article about PNG at http://www.hotwired.com/webmonkey/99/09/index0a.ht ml that goes into PNG's history and it's failure to penetrate the web. They say that IE mac doesn't support it yet.
Of course, it's the old chicken and egg thing. Part of the problem is that a lot of people haven't updated their browsers. In fact, I still stumble across some old netscape gold (16-bit, even on Windows 95 machines!) and IE 2.0 (back from when it was part of "Microsoft Plus!") browsers here where I work.
You know what better compression will mean in the face of the big pipes 'Just around the corner'? It means NOTHING. When I can download 100 MB in 10 minutes and have a 100 GB drive, do I care that a .wav file takes up 80-120 MB? No. Hell with the proprietary MP3 format. Hell with all those proprietary ass raped monkies.
slice some bread with our new Y2K compliant bread slicer and you'll survive
I do know about these Summus people though.
I think that's enough to change my mind. I don't want to give them an excuse to spam me every time they release an update to their ActiveX control. Ugh.
The "cue the foo posts in 3, 2, 1..." posts will commence with no subsequent foo posts in 3, 2, 1...
oops, bye-bye features.
I guess I can buy a multifractal view of wavelets.
But, as I've noted in my comments elsewhere, there are many interesting classes of images where one might use (ugh) fractals as some sort of descriptor, but wavelet based compression doesn't work.
These demo photos are often carefully selected. They
contain lots of fine detail that are compressed well with
the particular wavelets that are used.
Wavelet compression has been around for quite a while. It is also used in digital image processing. I know that doesn't answer the question. But if WI is going to be a proprietary format, it wouldn't take much to create an open source format that uses Wavelet.
As I understand it, Wavelet is the same as FFT except the basis function is different.
..out of the box compression will drop it.
:)
Properly selected one will not, and will be more efficient than Fourier, MA or other denoising technique. It is also very efficient for automating your analysis. Think machine vision.
I use wavelet transform too search for some features in time series data. Works excellent.
Unlike windowed FT preserve important singularities.
WT is a broad subject, and it seemed to me you picked up one particular implementation that is not up to your goal. E-mail me if you want to discuss this problem, I am always interested in new applications
<^>_<(ô ô)>_<^>
I have been studying wavelet compression for a number of years now.
/.'ers are interested.
It has been well proven that well coded wavelet transforms can be made to be more accurate than the DCT (Discrete Cosine Transforms) that are at the heart of both the JPEG and GIF formats.
The problem is, most of these formats are still at the academic stage of coding, have been licensed into extinction, and in general, don't offer the rest of the world much of a reason to come on board.
After spending a few minutes perusing the Summus Website and their technology, only one item struck me as being worthwhile -- the idea of focus regions, i.e., areas where less compression can be used to maintain higher fidelity to the original image.
Two things I would like to see:
1. this regioning technique incorporated into PNG, JPEG, etc.
2. a fast, Open Source wavelet transform which all of us Linux coders can put through the grinder until it is as worthwhile as JPEG and GIF.
I am willing to put up the server space for an OSS wavelet project if enough
Feel free to send an inquiry, but make sure you indicate both your coding and wavelet experience in the body of the e-mail; items without both will simply be trashed.
...Open Source isn't the only answer -- but it's almost always a better value than the alternatives...
GIF has 8 bit color, PNG has 24 bits.... a lost battle for GIF.
I used to run one of those silly free Web graphics sites. I tried to put PNGs on it instead of GIF but all the sillies didn't like it because they didn't know what a PNG was. I also tried to release my graphics under the GPL (I know, it's for software, but what the hey) and I got flame mail from those copyright-nut artists. Argh, I guess I can't win.
I just have to know is
"----------------------- Chop 'em"
supposed to be like a line of coke or crystal meth or what?
The problem is not with people lacking upgrades, its the cost. It used to cost for netscape, leaving Internet Explorer as the option, but now both are free. :P.
We here in the UK have the major problem that to get the latest stuff as soon as it comes out we have to spend hours on the phone, and I have encountered a phone bill of around £20 just to download netscape for my Linux machine.
The other alternative is buying magazines, which now include browsers on them. PC Plus has also made a bold statment by doing a series of Linux articles and software releases on their CD's, which means we're much better off.
My point? You americans have it easy with free calls
Wavelets are hardly new - the idea has been around for at least 3 years. I don't recall where I first heard about them (Scientific American?) but AT&T had a large interest a couple of years ago. In fact, a quick search for "wavelet" on altavista turned up wavelet.org, which is sponsored by Lucent Technologies. Other pages turned up included papers written about wavelets, and wavelet related source code. As I write this, I am downloading a package called wavelet.0.3.tar.gz - it is essentially a wavelet construction kit (grayscale only). The file is dated 1/29/97. It's about 700k, and there is no copyright notice / license on the page I'm downloading it from, nor in the source code.
So what we have is an old image format which hasn't caught on yet. I find it hard to believe that a community like the one here at Slashdot has never heard of this before, much less played with the freely available source code.
PNG is a lossless compression format. It's similar to GIF, but patentless and can handle more than 256 colours (among other things). This makes it good for non-natural images, especially those that suffer under the hands of lossy compression formats like JPEG. However, with lossless compression, your're seldom gonna get the same kind of compression ratio, simply because (rather simply put) lossy compression algorithms throw away a lot of the data, so they don't actually have so much to compress.
/. readership in general, you use your brain and make sure you do actually know a little about what you're talking about.
So, I suggest in the future before you start insulting the
cheers,
Tim
> Have you tried to surf todays web using
> Netscape or IE 2? Half the web sites do not
> work, and a lot of the ones that do look
> terrible. People will upgrade their browsers if
> the web becomes unworkable with their current
> one.
That's just the thing, they don't. *You* know that the web isn't supposed to look like that, but they don't... Not to mention, there are plenty of people like my mom who have a 386 just for word an a little browsing to check the weather... try putting Netscape 4.x or the IE4 beastie on it. Watch it crawl.
> How many people do you see still using Mosaic
> because they are affraid Netscape is buggy?
Do you run a web site that gets a decent amount of hits? Check your logs, you'd be amazed.
:)
WWJD? JWRTFM!!!
Anyone care to make a nice GIMP plug-in for this?
No compression algorithm will work on all images. And, for any given method, there are many types of important images that can't be compressed.
is that they are proprietary. They can't ever hope to make any real money off that. Who in their right minds would use a WI image on thier web site.
Nuff said.
--
Matt. Want XML + Apache + Stylesheets? Get AxKit.
Hey: that perfectly describes the JPEG software I have. Mind you, not _all_ jpeg will do that- I'm talking Boxtop Software 'ProJPEG'. It takes noticably longer than generic JPEG to run, because it's doing things like (if I remember correctly) optimizing huffman coding of whatever the hell it's doing... *boggle* at any rate, the 'curve' of this is striking. It's not that much greater than, say, Photoshop's version, for super high quality that's supposed to look lossless. Still looks like a jpeg, maybe a bit smaller. However, when you start getting ruthless, look out! I think my limit was nearer to 400:1. This looks ugly and it's more usable to go with between 30:1 and 90:1...
Examples? Easy: my art pages (which also include a bunch of linux tiles and titlebars in XPM) have background pictures that are JPEG. They are all 1024x768 and are around 100K in size. They are here. (And I should give my mac/linux dualboot box up and start doing everything in Windows for what, exactly, mister proprietary compression vendor sir? Feh)
They claim to be the real China, right?
wavelet compression has many advantages over jpg:
no 8x8 tiles, progressive loading (with only few percent of the whole data astonishingly good picture quality is possible)
wavelets are very well suited for compressing natural photographs (I have seen some demos)
I know Netscape 4.x has support for them the only problem is I'm not sure if it has total support for them and their animate gif killer MNGs, or whatever. PNGs can support alpha blending on every possible color and well as full transparency, but you don't see too many view and even few editors that use all of these features.
We probably will be using jpeg and gif for quite a while. I remember the first time I heard about png, it was supposed to be this revolutionary format that would be used on the web and everywhere. And now where is png? Well, it's gnome's preferred gfx format, if that counts for anything :)
MoNsTeR
The best idea they could think of for this was webpages? And after they had this brainstorm they thought ActiveX objects and plugins would suffice?
Prediction: We'll never hear about WI again.
The reason I use GIF's instead of PNG's for my lecture notes is for the animations ... to be able to animate the little market diagrams we economists pester students with.
I haven't seen any PNG animation tools, but is animation part of the PNG standard?
Re:
http://www.hotwired.com/webmonkey/99/09/index0a_p
doesn't everyone here remembering hearing about this from either slashdot or freshmeat... .. i'm thinking it is more like one year though..
this is at least 6mos old
Marques Johansson
1) JPEG technology is a decade old and not cutting
edge in image compression. Creating a standard
for anything takes a long time. But there's
JPEG2000 ahead
2) PNG is not lossy and cannot be compared to JPEG.
3) Good proprietary wavelet-based libs are already
available, also with region-focus (Pegasus Imaging
do this, www.jpg.com).
4) BTW: the article says
I images aren't just smaller than JPEGs, they also preserve the original image better
That's nonsense, lossy algorithms can shrink images to any size, it's about how good images still look at a given bitrate, so they're not smaller, they look better for a given bitrate.
--
ms
Maybe a few hundred pieces of email asking the same quetion will pound them into submission. :-) If anybody can get through to their web site, post an address.
The "cue the foo posts in 3, 2, 1..." posts will commence with no subsequent foo posts in 3, 2, 1...
Did anybody else check out the section on the Netscape plugins? They support Win95/98/NT 4.0 exclusively, with NO mention of ever supporting ANY other OS. Considering that it is a proprietary scheme, this is probably not much of a problem. Hope it doesn't catch on...
Sorry but PNG is cool. You might not use it but
it is a free and open format. Do you really
want to put up with that GIF patent crap?
Use PNG today!
Common, who with any amount of brains is gonna use a proprietary format, let alone a proprietary IMAGE format. Yeah, how well is this thing going to last, and how well will it ever be supported. It's common law that proprietary formats always die. There is of course some exeptions, like the M$ Word format, but that's only used a.) because Word has the monopoly on word processors, and b.) also intelligent people only use it (forgetting the fact that an intelligent person would use LaTeX :) for inside there own editor, not for when they want large numbers of people to see it. No large company has a monopoly on image formats, Free Software has a stake in it, many smaller companies do, and so do large companies like Adobe, Macromedia (i really hate proprietary software, but that company comes out with some really awesome software), M$, and Netscape.
Check out TrueSpectra's products. Resolution independent images can be used instead, although they use a dedicated server to do it.
http://www.truespectra.com
Use png damn it, god i hate how theres so many
lamers on this planet!!!
IE4 supports png for ALL INSTALLATIONS
Netscape with quicktime plugin supports PNG
and should do with next version, and it might in
4.5 i think....
So why not use it???? coz its still not 100% penetration.... and too many people wont use it.
Until netscape get of their ass! and finish v5.0 , thats it... butt at least ie4 does it, so theres no problems!!!
How do I use Netscape's native support? On my WinNT (ritual spit towards Redmond) box I installed Apple's Quicktime 3 plugin, which takes over the PNG format (and a boatload of others, as well). I'm running Netscape 4.5. How to I tell it to disable the QT plugin for PNGs and render them natively? This is somewhat important to me, as I'm doing some web development with PNGs and I want to see how Netscape handles them. My other recourse is to get another machine in here with a "clean" install.
I looked at the "Applications" entry in the Preferences dialog. "PNG" is listed as being supported by a plugin, but the "Edit" and "Removed" buttons are greyed out. I can't find a way to change it.
An oddity in the Linux (hail Linus!) version of Netscape 4.5: I have no plugins at all installed, and the "Applications" entry in Preferences tells me "PNG file -- Unknown: Prompt User". Netscape does render PNGs, though, even though it claims not to. But it seems that PNG is not its preferred format. On the PNG home page the PNG logo is given in an OBJECT tag, along with a GIF. Netscape renders the GIF instead of the PNG.
Chelloveck
I give up on debugging. From now on, SIGSEGV is a feature.
This is an interesting metric. Hmm, is there a reference site that you recommend for comparing
different lossy compression schemes using ROC?
OK, you buy a sense of humor at K-mart, now ...
lake effect weblog
{Network engineer in Chicago--looking for work!}
yup, it's all in the model. And, that's the hard part. Wavelets, et al., are too generalized for many classes of interesting images. They may work well on some classes, as you note.
I know nobody knows about PNG, but it works, works well, and I'm quite happy with it thank you very much. Plus, Netscape at least v4.5 supports it "natively" and of course so will Mozilla.
It lets the innovative slip in under the "radar" of the entrenched constituencies. By the time the "old boys" figure it out, they're history (or we at least have enough of their market to make us happy).
Besides, a friend of mine, one of those euphamistic "high government officials" (in the sense of high grade, not in the Bill Clinton sense of not inhaling Monica) once noted that "if you have no enemies, you haven't done your job."
So, lets have more NIH from all the back room boys! Meanwhile, we'll be demonstrating working systems to your customers.
Posted by Charles Bronson:
o toshop.html
According to Summus' page, the Photoshop Plugin alone costs $150. I see this format going... nowhere. Surprise.
See for yourself at: http://summus.com/products/4u2c/photoshop/4u2c_ph
"Summus' 4U2C(TM) Adobe PhotoShop Plug-in extends Adobe PhotoShop's file formats to include 4U2C(TM) Image Compression. This Plug-in allows PhotoShop users to view Summus Wavelet Images and convert other image formats to Summus Wavelet Images. Summus Wavelet Images file size is controlled by File Size, Compression Ratio or Image Quality.
Price (US and Canada only)
$149.00 + S&H ( SC residents add 5% sales tax )
For International Pricing call or email"
Netscape supports PNG, and I believe that IE does as well. I've seen web pages with narry a GIF in sight, all images having been converted to PNG.
Elevators smell different to midgets. ProZac
Of course, I'm still waiting on proper support for PNG graphics.
Weblogging Considered Harmful:
Wavelet compression is not entirely unrelated to fractal compression. Like the fractal approach the artefacts are more in sympathy with the way the brain works than are the artefacts of 2D Fourier based approaches, like JPEG. It is not, therefore, entirely incorrect to say that files are both smaller and better. You can make the wavelet file smaller than a JPEG file before a mathematical measurement says the loss is similar. However, when you look at the results the wavelet picture still looks considerably better, as the artefacts are more in sympathy with the way our eyes work.
Wavelet and fractal compressors have been around for at least a decade, for audio and other data as well as for images. They have previously been highly asymmetric in their compute requirements - they need huge compute power for compression, but very little for decompression. This has previously made them impractical for applications like mini-DV cam-corders - the key reason MPEG-2 is still 2-D Fourier based. Tricks for getting the compute requirements down are now coming through. The next MPEG standard may change to wavelets, and I understand the next JPEG standard definitely will.
I'll pipe up here... I use PNG's for my Family History Project (genealogy). The idea is to scan the photos we have now at a highest quality (either 300 or 600 dpi, depending on the age of the photos... new photos suck for resolution, gotta go back about 50 years for the really good stuff), and store them lossless in the smallest size possible! PNG is perfect for this, I have about 6 CD's of these PNGs to deal with.. Yes, 3 gigs of family photos (about 600 or so total). However, I convert to much smaller jpegs before I go online.
Should I dare a minor slashdot effect? It's at http://genealogy.arbutus.cx/. There's some cool old photos there.
Discrete Wavelet Transforms are chosen as the transform for JPEG 2000, so we will get the best of both worlds.
/juels
You don't give a rat's rectum now, but you'll be bitching like hell when your pointy-haired-Dilbert-boss comes to you and says, "We've opened a branch office in Podunk, and they need to do e-commerce. Put that catalog you've been working on online."
You *should* give a flying fluck about web graphics, because it all seems to be converging. It's no longer the technology that leaves boys typing with one hand-- it's the tech that my mom just ordered a car with.
That's a whole bunch of images according to my calculator (1. * 10^1233
Okay, I'll use all PNGs on my next web project. Then I'll go cash in my unemployment the next week. I'd love to use PNG, but it's kinda hard to go to a client who's hired my company for a project and explain: "Well, no, your audience will probably be limited to the techie 5% of the Internet population, but it's the Right Thing to Do."
It's just not that simple. I really wish it were. If it were, I'd have nary a speck of WinNT running in my local world.
Make an image of random numbers. Put a faint line in it.
Now, compress that image with JPEG, LZW, or wavelets.
Upon decompression -- the line is either gone or you get a negative compression ratio.
Sound far fetched? The image described above is a good proxy for things like radar images of the ocean.
Actually, PNG is the internal image format used
by MS Office. It is not widely advertised, but
acknowleged. libpng (reference implementation
is available in source) The only issue I have
with it is that it relies heavily on FP hardware.
A truly portable implementation should work on
integer-only hardware.
> And I believe its at least a year old now.
Try closer to three years old.
(AC, Follow up from last night)
/'er offered to sponsor?
Thanks for the follow up, SLEW. As I was reading your comments I remembered more of the article about GIF's, JPEG, etc. -- comparing their respective algorithms as opposed to connecting them.
A last question: do you think there is any merit to the idea of an Open Source wavelet compression algorithm like the one
Incidentally, "focussing" is also possible with JPEG images. And, furthermore, wavelet compression isn't state of the art anymore either.
How Damn Much for this format? AC
i did a little bit of web research and find that their is stuff about wavelet dating back to 1990... :)
i also find www.wavelet.org
on there second newsletter they give c++ source for the compression codec...
Marques Johansson
I used their Windows and Mac versions at my previous company. If anything, this will end up being a "Windows" format. Their Mac version was horrible - and their Mac programmer was pretty clueless as I recall.
It was slow as well...
I read an article about some fractal compression which compressed an image to 9k where JPEG took 109K. That was 8 years ago. Never heard anything about it since.
I guess that this new compression will prolly go the same way.
i don't know about this particular algorithm but :)
i've worked with wavelet image processing before
and it's fucking incredible. you can isolate the
singularities in a signal and scale the amount
of information used to encode it accordingly. it
can compress 2X better than fourier (jpeg) with
the most braindead implementation. i wouldn't be
surprised to see 4x better compression in the
next couple years... now to work on mp4
> there's just one global compression table, which means that the different files
.tar.gz of BMPs would probably be smaller than
>in the archive are all compressed against each other.
Which is EXACTLY why it's a BAD idea. More files=more diverse data=less efficiency in the compression table.
>A
>a lot of single GIFs, maybe even smaller than a
>.tar.gz of GIFs.
You should try your experiment and see what happens.
PNG is out there and it's being used extensively in environments where the GIF generating tools just don't cut it. It's supported everywhere and is free from the stupidity associated with GIF compression patents.
Where is it now? How hard are you looking?
Richard
I create 2D and 3D images and animations for a living. PNG really saves space on my computer by allowing me gif-type compression at 24-bit color. This format has been used in the 3D world for several years now... I expect it to replace GIF sometime in the near future.
Michael C. Hollinger
Naturally, this will depend severely on things like the size of the images.
I tried this experiment with 2400+ small bmps we have here (average size, 1k or so). The tar.gz was about 1/20th the size of the directory of GIFs. This won't always be the case, I'd guess. Since the images were so small, the GIF directory probably had a higher proportion of redundant info (ie, 2400+ headers) than a normal set of images.
For the PhotoShop plugin. Not worth it, especially since the web browser either needs a plugin or an active-x component.
I'll stick to JPG & PNG, Corel PhotoPaint supports it natively, it can be viewed on Linux, Windows, Apple, etc. I refuse to use an image format on my website that would make it readable only by windows users.
if your mom ordered a car only having seen a picture you should both be shot
Nope, not quite. You look at the pictures, then order someone to come by for a test drive. What, you think those auto sites just UPS it to you?
You are correct in the statement that GIF uses LZW, however, I believe the previous poster was correct in mentioning the Discrete Cosine Transform. I haven't looked at the GIF spec in years, but if I remember right the DCT was used technique to determine the optimum 256 colors to be included in the GIF palette in any non-fixed palette image.
Secondarily, the MPEG Quant scale is (as I understand it) mathematical, where the regioning shown on the web site appeared to be selectable under manual control.
As I have not worked with graphic files for some time, anybody out there care to offer clarifications?
It might surprise everyone to know that wavelet compression may be coming to a browser near you sooner than you think. The next version of JPEG (JPEG 2000) will support wavelet compression, so assumedly this will also be incorporated into browsers.
Regarding the quality and use of wavelet compression: This type of compression is nothing new. Wavelet compression has been used for graphics and audio for some time now. While different wavelet algorithms perform with varying degrees of quality, a good algorithm will provide much higher quality and better compression than JPEG. Why? Read some technical papers on it - I'm not about to explain it here. I personally have seen it used not just for web browsers, but for field applications such as compressing medical images (CT/MR scans, etc.) Compression ratios of 100/1 are sometimes achieveable with little loss. So, no, it is not a waste of time or a fad.
I'll admit, it will be several years perhaps before JPEG 2000 is fully supported by all browsers, but if JPEG continues to be the standard, we will have wavelet compression widely used in a variety of applications.
Those of you who are naysayers, do a bit of research on JPEG 2000 and you may change your mind. Otherwise, don't complain.
Yes there was bogus claims/or bogus images. Fractal compression is famous for the "graduate student algorithm" (lock a student in a room until he tweaks the compression/subdivide the image,... to an appropriate compression level). The results may be probably better than JPEG for automated methods and normal images, but since everything patented, proprietary, and locked...
Most problems with ROC can be fixed with a compander or in the general case, histogram
renormalization.
Although I personally think wavelets are a lost cause in the generic image compression arena,
people have used them successfully in specific areas where the data has known characteristics
which are supported by models (e.g., fingerprints, synthetic aperature sub-millimeter radar).
Wavelets also have many uses in analysis.
The DCT and most wavelet transforms are perfectly able to represent any images since they are
non-singular transforms. The compression artifacts are not because of the basis
functions, but the quantization of the coefficients. Most quantization algorthims are
naive so they produce naive results.
That's why I don't think wavelets have a chance. Wavelet people are so wound up in producing better
wavelet functions, that they end up ignoring quantization and companding improvements where
the DCT guys stopped playing around with the basis functions a long time ago and are years ahead in
the quantization, companding, and entropy coding areas.
BTW: It is amazing how people dis algorithms using the NIH (not invented here) metric.
The only thing that's going to change what gets used is people actually changing the formats. If you don't want to see more gif's, use png's for all your pages. Most browsers support them (well, IE 4, Netscape 4 and Mozilla, plus any browser that uses external image softwear.)