LibTiff Mailing List
TIFF and LibTiff Mailing List Archive
February 2008
Previous Thread
Next Thread
Previous by Thread
Next by Thread
Previous by Date
Next by Date
Contact
The TIFF Mailing List Homepage
This list is run by Frank Warmerdam
Archive maintained by AWare Systems
Thread
2008.02.18 13:12 "Re: bug? JPEGQUALITY seems not to be read correctly", by Joris Van Damme
John, jcupitt@gmail.com wrote: > On 18/02/2008, Gerben Vos <Gerben@zylab.com> wrote: > > Some rather sophisticated code is available on the Internet to > > reconstruct the quality value from the tables in a JPEG. > > Another approach might be to make TIFFTAG_JPEGQUALITY into a real tag. > This could be a simpler solution, and almost as good. > > I noticed that TTN2 (the revised JPEG-in-tiff note) obsoleted tags 512 > -- 521. Perhaps one of these could be reused? RESTART interval, > perhaps (it's a TIFF_SHORT): Reusing obsoleted tags for new purposes isn't a legit approach. New purposes, requires new tags, to be newly allocated from Adobe. > Old-style jpeg-in-tiff would be unaltered. Of course Q settings are > not transferrable between jpeg implementations, but it sounds like the > Q guessers are not exact either. It seems to me that saving the Q > setting as a hint would not be any worse in practice than adding a Q > guesser and would be far simpler (just a few lines of code). One point that makes this whole argument moat is that this LibJpeg quality setting we're all familiar with, ranging from 0 to 100 is itself a LibJpeg specific measure. Other JPEG codec implementations might use a quality parameter with three different settings, or five, or whatever. Some implementations don't reflect quality as a direct scaling of quanitzation tables, but have specific experimentally derived optimal quantization tables for a more limited range of quality settings. Defining a tag that reflects a measure very specific to the LibJpeg implementation is dodgy, at best. Another thing is, there's a number of other things that influence quality, aside from the one you all seem to be focussing on which is quantization table scaling. There's also, for example, rounding issues that can become significant in fast small-bit-sized integer approaches that sit well with SIMD, measured against double precision floating point reference approaches. And last but not least, there's the quality and file size impact of subsampling, in TIFF speak that could range from [1,1] to [4,4] which does make a very substantial impact. This already hints there's two more fundamental questions to ask, here. What is 'quality'? And last but certainly not least, why should we care to make any guess or preservation? Could the need in some cases perhaps be an indication of some wrong thinking somewhere else along the process path? Best regards, Joris Van Damme info@awaresystems.be http://www.awaresystems.be/ Download your free TIFF tag viewer for windows here: http://www.awaresystems.be/imaging/tiff/astifftagviewer.html