In the late 1990s, one of the hot technological topics in imaging was about the use of wavelet transformations to optimize JPEG’s somewhat outdated compression algorithm, in order to save considerably more space while not sacrificing any quality. At the time, numerous vendors came forward with solutions in this area, all bolting on proprietary extensions to JPEG, making the resulting file decidedly everything but JPEG-formatted. A standard had to be created, or so people thought. And thus, JPEG 2000 was developed by the Joint Photographic Experts Group group. The legitimate successor to their own JPEG, JPEG 2000 not only introduced wavelet, but – amongst others – also a lossless compression option, error checking, support for an alpha channel and, curiously enough, incompatibility with EXIF, which is somewhat mitigated by the new file formats JP2 and JPX, which have their own flexible means of metadata storage.
About a decade later, it’s fair to say that JPEG 2000 has yet to make any impact, so perhaps it never will. I vaguely remember a conversation a few years ago with a heavy proponent of JPEG 2000 who wanted to help implement the format in the Mozilla project, to which I contributed at the time. The effort not only went nowhere; it was also followed by one in the opposite direction: Mozilla, at the time the one major piece of software to support JNG and MNG, dropped support for the two for lack of adoption. (KDE apparently has native JNG and MNG support, but hey, I said ‘major’.)
What is it about JPEG 2000 that makes it not worth adopting? Open source developers will be quick to point to the legal complexity due to wavelet-related software patents, as Wikipedia briefly explains. That’s definitely one big obstacle, but I don’t believe it to be the main reason.
Back to the late 90s: normal hard drive sizes were in the single-figure Gigabytes, floppy drives typically still shipped with computers, and DVD burning at home, USB flash drives or external USB- or FireWire-connected hard drives were completely unheard of. In fact, USB just started catching up, and FireWire was extremely rare. Your average Internet connection was narrowband dial-up, or ISDN at best. Digital cameras, which were not exactly commonplace for consumers, had storage in the area of the low Megabytes, and usually connected via a serial cable – printer port on the Mac, COM port on Windows. If you get my drift, storage of photography was a wholly different topic than it is now, and few would even have dared to utter the space-guzzling letters “RAW” as a non-compromise-quality format. A wavelet enhancement to JPEG seemed necessary, not just a nice bonus to itch out some more space. Finally, the lack of EXIF at the time made metadata storage hard, calling for an answer.
These days? Not so much.
Computer hard drives have hundreds of Gigabytes of space, you burn to DVDs (with HD-DVDs and Blu-Rays coming up soon) and backup to your USB stick or USB/FireWire/eSATA/SAS hard drive, an Internet downstream of 512 kbits/sec is considered low-end, and multiple-Gigabyte memory cards for digital cameras can be had for two-figure amounts of money. For higher-tier amateur or professional photography, RAW is largely considered essential, with tools increasingly revolving around its use. A wavelet enhancement to JPEG suddenly seems like an unacceptable waste of the time and intellect it takes to develop such technology.
Mac OS X ships with JPEG 2000, which is nice when you need it, until you realize you really never do. I do not recall a single image that I wanted to look at that was in JPEG 2000 and wasn’t a demo image for the format. Nobody cares about JPEG 2000. Its main touted advantage is its better compression, but the advantage only becomes remotely relevant once you lower the quality level to the point where it becomes unacceptable for anything but a quick overview of the image, rather than a decent representation. While JPEG still leaves us with no facility for alpha, PNG has moved in to fill that void almost entirely, as lossless image representation seems more fitting in most such contexts anyway. And EXIF, which ironically is incompatible with JPEG 2000, makes its metadata enhancements so superfluous that any remotely useful digital camera fills its fields extensively.
Idealists like you and me may prefer if commonly-used formats are exactly as efficient as they could be, scratching out every last resource in our modern computer hardware and software. But moving from one format to another is always coupled with massive costs of migrations and incompatibilities. Web browsers have been particularly slow to adopt new image formats, which is rather pathetic, but cannot be attributed to any vendor in particular; they, too, will play the ‘good enough status quo’ card, and rightly so.
Other arguments against JPEG 2000 have been made, most prominently the increased encoding and decoding complexity, requiring major improvements in the efficiency of digital camera chipsets, and possibly battery life losses (and who would want that?). I also hear that the actual implementation is a major pain, as is frequently the case when formats are specced not by those who have to end up implementing them, but those who merely have a good degree and want to to account for the near-impossible exception case #8514.
Inevitably, something like this comes up on reddit and such every now and then. Don’t mind them, though; they’re just being purists, and there’s nothing wrong with that.
Others' Thoughts
Comment on February 20th, 2007 at 11:21 am
It can be worthwile to use lossless JPEG2000 compression inside PDF files. Unfortunately, it requires Adobe Reader 6 (or anything else that can read PDF 1.5), but that’s not really a major obstacle anymore.
Other than that, I don’t see much practical use, no.
Comment on February 20th, 2007 at 11:42 am
Is lossless JPEG2000 that superior to PNG?
http://tim.oreilly.com/cs/user/view/cs_msg/27918 says:
Wikipedia’s comparison section similarly states:
I wasn’t aware of this. Does this still apply even when a PNG is optimized through pngcrush, optipng and the likes, i.e. is this benefit inherent in the lossless algorithm(s) of JPEG2000?
Comment on February 20th, 2007 at 1:33 pm
I think it boils down to the fact the regular jpeg is good enough for most people and most people don’t like change unless their is a very compelling reason to do so. I changed from Internet Explorer to Firefox (on Windows) because of all the negatives associated with Explorer and positives I couldn’t get in Explorer at the time. I also changed my habit of using the finder and dock to using Quicksilver because the short time to learn it saves a lot more time in the long run. I would have probably used JPEG 2000 if it was just me and we were int he pre-Internet days (as most people are concerned) but get others to adopt to something is a bigger challenge.
Comment on February 20th, 2007 at 3:54 pm
For a typical photo, I don’t think PNG can beat JPEG2000 on any setting. So yes, I would say it’s an inherent benefit of JPEG2000. Still, I think PNG manages to do quite well considering its intended field of application.
Comment on February 21st, 2007 at 2:58 pm
JPEG 2000 has made its way deeply into medical imaging applications. It has been included in the DICOM medical imaging standard, in part due to a number of positive Receiver Operator Characteristic studies designed to assess the format’s diagnostic accuracy for radioligical imaging applications.
Have a look at NLM’s PubMed: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?CMD=search&DB=pubmed
using the following search string: jpeg2000 OR “JPEG 2000″
++++
To address the “no one’s using JPEG 2000″ assertion variously made by yourself and many others on the Web, have a look at the Library of Congress’ “Sustainability of Digital Formats” webpage on JPEG 2000:
http://www.digitalpreservation.gov/formats/fdd/fdd000143.shtml
Note especially the comments on the evolving adoption landscape in the Cultural Heritage community. Large quantities of content of specified image quality must be managed and delivered to end-users who may not want/need the full-quality image.
A number of libraries and archival agencies have been evaluating JPEG 2000 for the last six years (see http://j2karclib.info for more on this), and are presently putting J2K to good use in several heavy-duty applications like the National Digital Newspaper Program (NDNP):
http://www.loc.gov/ndnp/
See also a website that delivers content from JPEG 2000 files:
http://charlesolson.uconn.edu/WorksintheCollection/MelvilleProject/about.htm
The J2K files are self-documenting in they they have embedded in them catalog records and finding aids for the collection displayed.
Harvard University has made the Big Step and has designated JPEG 2000 as an official archival master file format that rests alongside the venerable but less flexible TIFF. Also, the upcoming IS&T Archiving conference in Washington DC will host several papers (for the 3rd year in a row) on using J2K for archiving & access purposes:
http://www.imaging.org/conferences/archiving2007/program.cfm
+++
FYI: A commercially available product line for JPEG 2000 enabled medical imaging applications may be glimpsed at:
http://www.aware.com/products/compression/jpeg2000_med.html
+++ FYI: I have commented on this topic at a number of other venues +++
Comment on February 21st, 2007 at 3:26 pm
A lossy format is useless for archiving and especially for medical imaging, so I can only assume that you’re talking purely about lossless JPEG 2000, in which case I’m saddened to hear that the your cited facilities are so happy to use a heavily patented, non-widespread format over a patent-free, very widespread one, just for the sake of saving roughly one quarter of storage size, and only in the case of particular kinds of images.
Wouldn’t such money be much better invested in improving PNG algorithms?
Comment on February 22nd, 2007 at 4:08 pm
The digital formats that are designated as “archival” reach that status by convention, i.e., a general understanding that the format’s capabilities and disadvantages can be managed by the archive operators.
Paper is tearable, flammable, and edible if you are a termite. It disintegrates and/or becomes moldy when it is wet (viz Katrina aftermaths for libraries at http://www.ala.org/ala/cro/katrina/katrinadamage.htm), and is bound to a single physical location.
In spite of this, paper can become an archival format when it is placed in an environment where the above threats are naturally lacking, or in one where they have been deliberately or inadvertently engineered away.
Harvard’s archival designation signals both their ability to manage the format from an IT perspective as well as their understanding of the format’s characteristics – including the IP entaglements characteristic of JPEG, MPEG, etc.
The PubMed articles are all about using lossy images. Given the quanitity of imaging that takes place in Radiology departments, they sought alternatives to (a.) photographic film and (b.) lossless digital imagery. This is why they wanted to establish that the diagnostic judgments made by radiologists would not be affected by compression effects. Some papers cited experiments with 10:1 compression ratios and beyond.
What is interesting about the analysis is that it is the same one they use(d) to tune X-ray films and related techniques. How fast a film should be, what kind of amplification screen should be used – all of those analog imaging parameters were set by this method.
One published study with satellite images – the improved acquisition and management of which was one of the original impetus behind developing JPEG 2000 – indicated that while wavelet encoding was statistically detectable in the images, there was no significant difference in the automated classification results.
Given an interest in knowing just how well or badly an image represents the original item, and what the effects are on a viewer’s evaluation of the original via the image of it, I am looking forwards to the appearance of similar studies using the other popular formats. Especially interesting would be ones using PNG and the proprietary Windows HD.
The evaluation of a format at the signal processing and signal detection level (which is what the ROC and image classification studies mentioned above do) that provides us with very targeted subjective consequences. One could essentially dial down image quality levels in any format using whatever means the format allows and see what the consequences are for real uses of the imagery.
Most people do not need that level of rigor in understanding their images, but where you are generating and managing tens of millions of images and still have to attend to image quality issues (the extreme case being the Digital Radiology one), applying this rigor is a good idea.
What we DO with these results, and what parties who have sufficient market power want to do with their favorite formats in spite of objective comparisons, however…
Comment on February 22nd, 2007 at 4:27 pm
But also by the guarantee or likelihood that that format will be usable (i.e., decodeable) in ten, twenty of fifty years. The probabilities of that are increased exponentially when a format is actually in common use, which is without argument not the case with JPEG 2000. Furthermore, due to the patents, the ability to write one’s own decoder in the worst case scenario that none is available is further limited by possible intellectual property claims from corporations that may or may not sympathize with the situation.
Paper is accepted by an archival format mainly because of sheer supply and common use, not to mention no need (aside from literacy skills) of decoding. No algorithm beyond what your eyes and your brain provide is needed to read the contents of a piece of paper. Any digital format is an order of a magnitude more complex than that, not to mention that in addition to the file format (e.g., JPEG 2000), you have to deal with the underlying file system (e.g., UDF, HFS+, ext3 and whatnot) and, of course, the storage medium, all of which can have their own intellectual property restrictions attached to themselves, and could potentially be rendered useless by computing technology changes. E.g., try reading a MOD song on an 5 1/4-inch disk with a C64 file system today, then consider that those were still in somewhat common use about one and a half decades ago, and now extrapolate what this means for the year 2022.
For what it’s worth, I wasn’t trying to bash JPEG 2000. It came up in a reddit post recently, with someone showing its lossy compression algorithm’s advantages, which without doubt do exist. However, they are for the time being completely impractical for an average user, for reasons I have stated above; mainly the lack of support in applications (both software and hardware), be it your average web browser or a consumer-level digital camera, but also such unfortunate developments as the incompatibility with EXIF.
If the three major browser engines out there (Gecko, WebKit and Trident) were to support it, or if one were to start and the other two to seemingly pick the idea up, I would likely use the format myself. Similarly, if my next digital camera were to support it, I would be interested in it as well. But neither is likely to happen any time soon, if ever.
Your applications are merely a niche*, and one that has too little attention among developers, the media or consumers for the format to have much of a chance at gaining momentum. I don’t think you have disproven that at all. What you have proven is that there are legitimate usage cases, which I never doubted to begin with; our disagreement is solely on their extent.
*) And by the way, I’m not belittling your work in the least. I am merely pointing out that it doesn’t represent a usage case for the masses.
Comment on February 24th, 2007 at 4:09 am
Actually, Second Life uses JPEG-2000 for its storage format on the backend, even though they prefer TGA files to be uploaded.
IIRC it was due to the streaming capabilities, as well as compression. They use the kakadu libraries as part of the non-open-source client.
–TSK
Your Own Thoughts
I'd love to hear your input. Just try to stick to a few rules:
Before you comment for the first time (or, after you have deleted cookies), you will have to answer a little challenge to prove that you are not a spammer.
Comments are written in Markdown.