View Full Version : jpeg2000 at the movies?
grumpy
1st September 2004, 02:24
(look here) (http://news.com.com/Digitizing+the+multiplex/2100-1025_3-5330706.html?tag=nefd.lede#yourtake)
A technology consortium called the Digital Cinemas Initiatives (DCI), created by the major Hollywood studios in early 2002, is finally nearing completion on a set of technical recommendations that is intended to rally the industry around a single technological standard. A few details remain to be completed, largely dealing with securing the files against unauthorized copying while in the theater. But the fundamental technology specifications, based on the JPEG 2000 video format, have now been chosen.
Doobie
1st September 2004, 05:30
Why JPEG2000? I thought that was a standard for individual pictures, not a video codec. What's the point?
Mug Funky
2nd September 2004, 02:37
i thought 3d wavelets were kinda given up on because they presented no clear advantage over DCT, but were more computationally intensive.
though, rududu codec is kinda cool, i wouldn't be archiving DVDs with it (and that's ol' fashioned block based MC, not 3d wavelet but 2d with moving blocks)
maybe it scales better for large frame sizes?
[edit]
*double-take*
it's only intra-coded! wow... they must have some hardcore network speed in mind to be able to play motion-jpeg2000 in 2k^2 res.
see here:
http://www.jpeg.org/jpeg2000/j2kpart3.html
Doobie
3rd September 2004, 03:37
JPEG2000 Part 3? Heheh. I figured that they'd have to just run a serious of pictures together. But, I would have thought that the loss of intraframe compression is too high a price.
Apparently they're not much concerned about bandwidth.
Wavelets naturally scale the picture size nicely. But, this is hardly important considering there are existing very good algorithms that will scale any picture nicely. Besides, how would this apply to movie theaters which should have standard resolution equipment.
If not for the loss of intraframe compression, JPEG2000 Part 3 would be good for video streaming where changes in bandwidth wouldn't stall the picture, just reduce the quality of some frames.
communist
7th September 2004, 11:57
AFAIK Sony already made a realtime JPEG2000 encoder that you can connect to their projectors (for Cinemas...).
Sirber
7th September 2004, 13:23
look the board for "rududu". it's a 3D wavelet codec.
reepa
8th September 2004, 19:57
I got interested in JPEG2000 after discovering a free encoder. I did a little test.
http://koti.mbnet.fi/vuonnala/riston/compare.png
JPEG2000 obviously eliminates the awful blocks, but it's not without its own artifacts. I don't know the math behind wavelets and compression but those JPEG2000 artifacts look awfully similar to bilinear resizing artifacts (diamonds). I was wondering if it was possible to increase the quality of jpeg2000 compressed images by using bicubic instead of bilinear resizing when decoding the picture (unless the problem comes with encoding or there is no resizing involved at all and the artifact similarity is just a coincidence :P).
Mug Funky
8th September 2004, 20:24
@sirber:
rududu isn't 3d wavelets... it's ME is block based and it's intras are wavelets. interesting combination, and it gave pretty nice results.
@reepa:
unfortunately that can't really be done - the wavelet transform is rather similar to having several layers of resized images sitting on top of each other.
if you changed the decoder to bicubic, the layers sitting on top of the low frequency ones (where we see diamonds) will be sitting at the wrong intensity levels - they were encoded with one resampling technique under the assumption that they would be reconstructed with the same technique's inverse.
there would be loads of artefacts from changing the resampling technique.
btw, what i say is mostly analogy - wavelets apparently are like a type of filterbank (rather like what mpeg-audio codecs use to convert a full spectrum into 32 sub bands), but with a few advantageous properties over the regular QMF filterbanks used by audio codecs.
(i hope somebody with mathematical grounding can explain this more accurately than i have).
it would be good if jpeg2000 used a different technique for it's wavelet decompositions, but it's too late to change the standard now, as it'd break all existing images.
vBulletin® v3.8.4, Copyright ©2000-2010, Jelsoft Enterprises Ltd.