4:4:4 and 10 bit

Logged In: YES
user_id=158189

Since the developers come from a broadcasting background
we are well aware of the need for high quality compression fro
professional environments.

We plan to support 4:2:0, 4:2:2, 4:1:1 and 4:4:4. In addition
we plan to support RGB and luminance only.

The number of bits per pixel is a more complex issue. Since,
at the moment, Dirac is a lossy compression system (not too
lossy we hope!) in 10 bit systems it is likely that the bottom
two bits from decompressed Dirac will not be the same as
those that go in except at very high bit rates. In this respect
the number of bits at the output is simply a matter of output
formatting.

On the input side it is generally better to take in as much
infromation as possible prior to compression to get best
results. How this information is used internally is subject to
further developement. Increasing the number of bits from 8 to
10, internally, may have a disproportionate impact on
performance whilst not yeilding any improvements in quality.

I think it is work bearing in mind that 10 bit professional video
was originally only using extra two bits, originally designated
spare, on the D parrallel digital interface. They were only used
becuase they were there. At that timethere was considerable
debate about the need for these extra bits. I think it is
generally agreed that when cascading processing these extra
bits are helpful. That said, processes like error feedback and
dynamic rounding mitigate the effects of only using 8 bits.

The bit stream syntax that the team and I are currently
drafting, and which we hope to release for comment very
soon, will include explicit support for 10 bit video.

Logged In: YES
user_id=820001

The 8bit vs xxbit debate has been a long one, especially
amongst Photoshop users. Some say there is no difference,
other say it mean life and death.

For me, I would be very happy to have 4:4:4 support before
higher bit depth, simply because it yields much better
improvement of quality, rendering at 24bits per pixel instead of
16 or 12 which is usual.

Most TV content isn't badly affected by the lower resolution
chroma channels of 4:2:0/4:2:2 because colors aren't very
saturated to begin with. However for CG and otherwise high-
hightrast and colorful content there is a clear difference.

An example I've made to illustrate: http://moment22.mine.nu/
yuv.png - remember this is with no compression. No matter
what bitrate you use this would be the best result.

Higher bit depth could certainly make a difference if the video
is to be post edited. otherwise we are limited to the output
resolution, which is 8bits per channel, unless using some
special hardware. The analog video-signal itself can probaly
contain much more detail than 256 levels...

If you want to post-edit image/video then you certainly can use
all the extra levels/bits. This is obviously visible in the digital
camera market where 12bits and more is standard.

Logged In: YES
user_id=820001

The 8bit vs xxbit debate has been a long one, especially
amongst Photoshop users. Some say there is no difference,
other say it mean life and death.

For me, I would be very happy to have 4:4:4 support before
higher bit depth, simply because it yields much better
improvement of quality, rendering at 24bits per pixel instead of
16 or 12 which is usual.

Most TV content isn't badly affected by the lower resolution
chroma channels of 4:2:0/4:2:2 because colors aren't very
saturated to begin with. However for CG and otherwise high-
hightrast and colorful content there is a clear difference.

An example I've made to illustrate: http://moment22.mine.nu/
yuv.png - remember this is with no compression. No matter
what bitrate you use this would be the best result.

Higher bit depth could certainly make a difference if the video
is to be post edited. otherwise we are limited to the output
resolution, which is 8bits per channel, unless using some
special hardware. The analog video-signal itself can probaly
contain much more detail than 256 levels...

If you want to post-edit image/video then you certainly can use
all the extra levels/bits. This is obviously visible in the digital
camera market where 12bits and more is standard.

Logged In: YES
user_id=820001

Hm, odd that there was a second copy of my post.

Anyway, what I was trying to say was that although 8bits is
not much for images when you post process, it should be
enough for video which is already pre-processed (perhaps at
higher bit depth).