[opendtv] Re: FW: Re: negative RGB (bit depths)
- From: "Tom McMahon" <TLM@xxxxxxxxxx>
- To: <opendtv@xxxxxxxxxxxxx>
- Date: Mon, 17 Jan 2005 17:25:51 -0800
One way to think about this is that these coders transform pixels into the
spatio-temporal motion estimation and frequency domains.
Having done that, there is no concept of pixel depth any more, and what the
entropy coders eats is transparently digested.
Greater pixel depth going in and carried thru the early stages of this encoding
process gives you more precision in some of the math
(and less repeated truncation/rounding), so the transforms produce less coding
noise and a more accurate motion estimation (and less
residue to code).
Keeping this extra precision in the encoder and decoder (but not in the
transmitted bitstream) costs you more in memory bandwidth
and memory size, but I think Moore's Law pays for that now, and the payoffs are
higher quality, potentially lower bitrate, improved
color space and less contouring, and better handling of over/undershoot. As
consumer displays break thru the 8 bit barrier (as
they've done already) these benefits will become more apparent, and whether we
like it or not bigger marketing numbers (N heads, 4X
oversampling, 24 bit audio) will become the norm.... So maybe at some point
we *will* see 10 bit emission (maybe on HD DVDs?).
As for PSNR measurements, I think we all know that is only a rough-cut on
quality metrics. It is also highly content dependent and
display device dependent. I've seen a lot of these tests.
-----Original Message-----
From: opendtv-bounce@xxxxxxxxxxxxx [mailto:opendtv-bounce@xxxxxxxxxxxxx] On
Behalf Of Ron Economos
Sent: Monday, January 17, 2005 3:19 PM
To: opendtv@xxxxxxxxxxxxx
Subject: [opendtv] Re: FW: Re: negative RGB (bit depths)
I read the paper last night. Pretty interesting, but given that the PSNR values
where 8 bit diverged from 10/12 bit were all above
45 to 50 dB, I wonder if this discovery is just a novelty. From my measurements
of HD MPEG-2 PSNR,
45 dB or greater PSNR is perceptually identical to the original (unless you're
from the planet Krypton), so it doesn't seem useful
to code at these bitrates (where the effect takes place) even for Digital
Cinema.
Ron
Tom Barry wrote:
>Tom McMahon was kind enough to send me the paper mentioned below, even
>though the list bounces attached .doc files.
>
>That paper does seem to suggest that feeding the AVC encoder more bit
>depth will almost always result in a higher (at least non-lower) PSNR
>for the same specified output bit rate.
>
>Or generally, a lower output bit rate to keep the same fixed output
>quality level when you input as 10 or 12 bits instead of 8.
>
>I'm just guessing this is because rate control will quantize down to
> a smaller number of bits / pixel anyway where needed. And for
>properly filtered material this may avoid some aliasing that would
>otherwise be created by the truncation to a lower bit depth in integer
>processing before reaching the encoder.
>
>But, whatever the reason, this is very good news for folks with > 8 bit
>encoders, which we all should go get immediately. I'm going to
>prematurely take this to mean we should ALWAYS be using 12 bits
>everywhere, and let the encoders make the choice based upon desired
>output bit rate and our quality trade-offs. Very cool.
>
>- Tom
>
>
>
----------------------------------------------------------------------
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.
----------------------------------------------------------------------
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:
