[Mp4-tech] DC precision in H.264??

Gary Sullivan garysull windows.microsoft.com
Fri Feb 25 11:44:08 ESTEDT 2005


Karthik et al,
H.264/MPEG-4 AVC includes some profiles for which the decoded video uses
more than 8 bits of precision per luma or chroma sample.  But I don't
think that's what you're talking about.  In decoded video, there is no
real distinction between "DC" and "AC".  Decoded video is in the spatial
domain, not the frequency domain.
What you're probably referring to is the way Intra DC quantization is
handled in H.261, MPEG-1, and especially MPEG-2 (and I think also in the
first version of H.263 and to some extent also in MPEG-4 part 2 and
perhaps in JPEG-1992).  I'm operating only from memory, so please bear
with me if I get some of the details wrong.
H.261 had a sort of a funny way of handling the quantization step size
for Intra DC coefficients that was different than the way it was done
for AC coefficients.  While the quantization step size for Intra AC
coefficients could be adjusted on a macroblock basis, the step size for
Intra DC quantization was fixed.  I think that aspect of the design was
basically copied into MPEG-1 without alteration.
In my opinion, there is no theoretical or rational justification for
such a design.  The Kuhn-Tucker/water-filling theory will tell you that
the DC and AC step sizes should all move together (maybe with some
amount of weighting of precision for perceptual reasons, but there is no
rationale for having a really huge step size for AC relative to DC or
vice versa).  One result of this is poor bit rate flexibility.  When
you're trying to reduce the bit rate, the fact that the DC step size
doesn't increase means that the AC step size must get huge to compensate
for that, so the picture can end up looking like a mosaic of flat 8x8
tiles.  The opposite is also true, when you're trying to improve the
fidelity, the fidelity of the Intra DC component never gets better, so
the resulting picture fidelity never gets as good as you would like it
to be.
During the design of MPEG-2, they noticed the second half of this
problem -- that even when they used a high bit rate, the locked step
size for the Intra DC component prevented the picture from getting as
good as it should have gotten.  To address that problem, they added an
alternative encoder choice of a different, smaller, locked Intra DC step
size.  The Intra DC step size would still be inflexible within the
sequence (or maybe it was within the picture), but it would be smaller.
This change basically fixed that problem, although it made the converse
problem worse (i.e., the AC step size must get bigger to compensate for
the DC step size being so small).  They called the modified version
"10-bit Intra DC precision", because the smaller locked step size
corresponded to using a 10 bit FLC to represent the value of the DC
coefficient (whereas in H.261 the DC component fidelity corresponded to
an 8-bit FLC).
The prediction of the Intra DC component was conducted in the frequency
domain in these older codecs, so when the quantization step size was
reduced to correspond to 10-bit fidelity, the precision used in the
prediction of the DC component was also increased to 10 bits to retain
consistency with the coded representation.
Version 2 of H.263 was the start of the movement away from such
inflexibility.  Its Annex I "advanced intra prediction" treated DC and
AC in a consistent fashion.
There were also some attempts to address the issue in the development of
MPEG-4 part 2, although it still ended up with a design that treated DC
and AC quite differently from each other (in my opinion, without proper
justification -- I and some others tried to argue that MPEG-4 part 2
intra prediction should operate like H.263 Annex I, but we failed to
convince the committee).
Throughout this evolution, Intra prediction remained in the transform
domain.
H.264/MPEG-4 AVC uses a different approach for Intra prediction.  It
performs its Intra prediction in the spatial domain rather than in the
transform domain.  This has various coding efficiency advantages.  It
also treats DC and AC in a very consistent manner -- when AC fidelity
gets bad the DC fidelity will get bad too, and vice versa.  The range of
supported quantization fidelity was also increased, so it is possible,
in a straightforward fashion, to attain higher picture quality with
H.264/AVC that could be achieved with MPEG-2 (and also possible to
easily make the quality lower than MPEG-2 to allow greater flexibility
of bit rate control).
Best Regards,
Gary Sullivan
+> -----Original Message-----
+> From: mp4-tech-bounces lists.mpegif.org 
+> [mailto:mp4-tech-bounces lists.mpegif.org] On Behalf Of karthik
+> Sent: Friday, February 25, 2024 12:06 AM
+> To: mp4-tech lists.mpegif.org
+> Subject: [Mp4-tech] DC precision in H.264??
+> 
+> Does H.264 support higher DC precision than 8 bits?Some sources have
+> large dynamic range and the range of many sources increases on
+> application of noise filters particularly.
+> 
+> Thanks in advance
+> 
+> karthiks
+> _______________________________________________
+> NOTE: Please use clear subject lines for your posts. Include 
+> [audio, [video], [systems], [general] or another 
+> apppropriate identifier to indicate the type of question you have.
+> 
+> Note: Conduct on the mailing list is subject to the 
+> Antitrust guidelines found at 
+> http://www.mpegif.org/public/documents/vault/mp-out-30042-Ant
+> itrust.php
+> 


More information about the Mp4-tech mailing list