JPEG’s Reply to Microsoft WMP: Better JPEG?
Posted by Sachin Garg on 15th June 2006 | Permanent Link
The new ITU-T Recommendation T.851 offers better (and faster) compression by introducing a new (backward incompatible) alternative Q15 arithmetic coding. Color precision is increased to maximum 16 bits per color component.
The change is only in the final lossless entropy coding stage. Thus current JPEGs can be losslessly encoded to use new algorithm and the lossiness of this new variant (the blockiness etc…) will be same as that of existing JPEGs.
Work on the new compression algorithm was started in 2004 by ITU-T Study Group 16. An alpha open source implementation is available.
Official results claim around 10% improvement in ratio over old JPEG. StuffIt’s JPEG recompression technology deserves a mention here, they claim a 30% better ratio by taking a similar approach. Thomas Richter posted some comparison results with both JPEG and JPEG2000.
Guido Vollbeding mentioned at comp.compression that Q15 is patent free. It removes some of the patented QM features (such as conditional exchange) and compensates for compression performance with use of an alternative probability estimation table (”Fast Attack”) and bit-stuffing technique instead of byte-stuffing (also regarding the latency issue).
There are some more developments/proposals due (like this one). A new extension ‘.jpa’ is also under consideration but they may stick with ‘.jpg’ like they did when progressive algorithms were introduced without changing file extension.
What I would like to see is a head-2-head comparison of old JPEG, new JPEG, JPEG2000, MS WMP and some of the other proprietary image formats, not just on compression ratio but also the relative ‘visual’ quality and speed.
Although we already have many image formats, only few were important enough to be bothered with. But now, the image formats domain is all set to become as crowded at audio/video formats. We have a mess of old/new open standards and a few proprietary ’standards’ have also showed up and will gain atleast some traction due to marketing muscle behind them.
Its going to be hard to say which format will win, but average users will now surely get more format choices and better technology (if they don’t go insane trying to find support for viewing their images in all the devices they use and all the software solutions they like). But I am not complaining, everything has a price :-)
June 15th, 2006 at 11:16 pm
A transcoding personal proxy (along the lines of my half-baked http://www.halfbakery.com/idea/Compression_20Proxy ) with a good plug-in system, and matching apache mods could greatly speed up the adoption of these new image formats. I wish people would care about image optimisation, particularly on the web.
June 19th, 2006 at 11:56 am
A transcoder is already available and they are atleast planning for integration in the popular IJG library.
The real question is if IE7 and Vista will support it.
August 2nd, 2006 at 6:30 am
Jpeg can be improved by using two simple additional processes.
1st. low bit rates can improved by using downsampling. Less artifact are visual if you resize the image to 70 % of the original and compress it using jpeg. You have to resize it back to the original size to see the results.
2nd. You can remove the artifact by using a postprocessing filter.
This one works very well.
http://www.compression.ru/video/deblocking/smartdeblocking_en.html
September 19th, 2006 at 6:24 am
http://www.cs.tut.fi/~foi/SA-DCT/results.html#ref_deblock - improves visual quality of jpeg images.
i think jpeg2000 isn’t superior to jpeg. it doesn’t suffer from blocking artifacts but usually removes much more detailt and introduces much more blurring. standard jpeg with right preprocessing (smoothing of luma and chroma planes and downsampling if necessary), optimized quantization tables, dct position dependent entropy coder and good postprocessing filter (like this presented above) gives much higher visual quality than jpeg2000 at usual bitrates (0.5 bpp and above).
the biggest advantage of jpeg2000 over standard jpeg is its scalability (jpeg2000 doesn’t need to downsample input data to compress better). the biggest disdvantage is poor performance - todays jpeg2000 coders are much slower (about 5x or so) and requires much more memory than baseline jpeg.
overall, jpeg2000 isn’t worth consideration. better improve baseline jpeg and use stuffit technology instead of simple huffman coder and old jpeg will be working fine.