[opendtv] Re: JPEG2000?

At 6:19 PM -0400 7/14/06, Manfredi, Albert E wrote:
>Cliff Benham wrote:
>
>>  The compression rate for JPEG2000 is incredible. An
>>  over 2 hour theatrical release takes only 250 G. All
>>  I frames, so no time fractured macro blocking.
>
>JPEG2000 uses a wavelet transform vs discrete cosine transform, and it
>offers a lot of flexibility, like scalability. But, as we saw on here in
>previous threads from time to time, it doesn't seem to be any slam dunk
>for compression efficiency, compared with alternatives. Or for
>complexity in encoding and decoding.

Compression efficiency must be assessed based upon the application. 
Clearly, it is unlikely that an INTRA-frame codec - designed to 
maintain pristine image quality - will produce bitstreams that are as 
compact as an INTER-frame codec designed for limited bandwidth 
emission channels.

As for complexity, intra-frame codecs typically use only a small 
fraction of the computational power needed for inter-frame codecs. 
This is true for the new chips that are being used for JPEG-2000 and 
other wavelet variants, just as it was true for the original JPEG 
chips, which were fast enough to allow their use for Motion-JPEG 
encoding of video at 30p/60i, more than a decade ago.

It is worth noting that analog devices developed a wavelet codec chip 
in the early '90s, but DCT based JPEG won the day in early non-linear 
editing systems. It is also worth noting that storage efficiency has 
grown so dramatically that it is commonplace for non-linear editing 
systems to use uncompressed source today, even for HD projects - a 
250 GB drive costs little more than $100 today.

>
>This suggests that its use for motion pictures would not result in
>better compression efficiency than M-JPEG, which of course is less good
>(in that regard) than MPEG-2 (H.262). And less good than H.264, as well.

There are certainly trade-offs when looking at the compression 
efficiency and delivered image quality of DCT-based M-JPEG versus 
wavelet-based JPEG 2000 and related codecs optimized for video/film. 
In my personal experience, these codecs are roughly comparable at 
similar data rates, however the compression artifacts of the DCT are 
more visible than the loss of detail that occurs with the wavelet 
transform. Hollywood has voted, and wavelets have won when it comes 
to theatrical exhibition.

M-JPEG is NOT less good that MPEG-2 when operating as an INTRA-frame 
codec - they are essentially identical. If you use some inter-frame 
compression it may be possible to produce a slightly more compact 
bitstream, at the cost of added computational complexity. Frankly, it 
is not worth it to add this complexity, given the cheap cost of 
storage. HDV is the exception to this rule; it uses inter-frame 
MPEG-2 compression to squeeze HD (rather poorly in my opinion) onto a 
25 Mbps tape-based recording format.

>
>Here's an article that compares the coding efficiency of various
>still-image codecs:
>
>http://jj2000.epfl.ch/jj_publications/papers/004.pdf
>
>The conclusions paragraphs say:
>
>"The results presented in previous sections show that new standards do
>not provide any truly substantial improvement in compression efficiency
>and are significantly more complex
>than JPEG, with the exception of JPEG-LS for lossless compression.
>However, from a functionality point of view JPEG 2000 is a true
>improvement, providing lossy and lossless compression, progressive and
>parseable bitstreams, error resilience, random access, region of
>interest and other features in one integrated algorithm.

Agreed. the compression efficiency is similar, but the added features 
are VERY significant. We will soon see these features exploited in 
many web applications, especially the region of interest  feature, 
which allows only the bits needed for more detail in the region of 
interest selected to be sent to the browser requesting the 
information.

>As to the 250 GB storage for a 2 hour movie, using JPEG2000. I would
>expect the result to be good for quality, and JPEG2000 is probably great
>as a future-proof digital format in archives. But compared with a 2 hour
>HDTV show, where HD uses every bit of the 19.39 Mb/s the entire time,
>the MPEG-2 file would be just 17.45 GB.

Apples and oranges. The article failed to mention that the resolution 
for the movie may be significantly higher than the 1920 x 1080 used 
for emission formats today. 4K x 2K is the goal here, and wavelets 
deal very efficiently with whatever information may exist at these 
higher frequencies.

>
>So you're talking about over 14X the size of the MPEG-2 file. Depending
>on how much greater the resolution of that JPEG2000 file was meant to
>be, a 14X increase may or may not be impressive.

I'll agree with Bert that there is nothing impressive about a 250GB 
movie file. We have understood these trade-offs and where each 
compression technology is best used for more than a decade.

Regards
Craig
 
 
----------------------------------------------------------------------
You can UNSUBSCRIBE from the OpenDTV list in two ways:

- Using the UNSUBSCRIBE command in your user configuration settings at 
FreeLists.org 

- By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word 
unsubscribe in the subject line.

Other related posts: