[opendtv] Re: Video compression artifacts and MPEG noise reduction

  • From: Craig Birkmaier <craig@xxxxxxxxx>
  • To: opendtv@xxxxxxxxxxxxx
  • Date: Wed, 25 Jun 2008 08:42:13 -0400

At 6:09 PM -0400 6/24/08, Manfredi, Albert E wrote:
Craig Birkmaier wrote:

 One thing is clear from the presentations, h.264 with the FRE
 is significantly more efficient than MPEG-2 ( 8 mbps for h.264
 versus 24 mbps for h.262 for essentially the same level of
 perceived quality.)

Honestly, though, isn't there some bone in your body that questions
these assertions, even after real world experience has shown that things
aren't quite so cut and dried?

No. These finding are well correlated with experience in the real world.

I saw that same presentation, and all I could conclude is that the
comparison was probably valid for A PARTICULAR MPEG-2 encoding. Or
perhaps, for some odd reason, the result only applied to a very specific
case of compressing 1080 at 24p material. So I did a little searching.

You see what you want to see.

One interesting tid bit I found was that MPEG-2 encoding can be much
more effective if the composition of the GOP is varied in real time,
according to the subject matter. And that decoders can handle this
variation quite nicely.

This is true for any encoder. But it requires some form of intelligence. For DVDs that intelligence comes from a compressionist who can hand massage the encoder setting to deal with difficult scenes. For a real time encoding without human assistance, one would need to create algorithms that detect various aspects of the content (scene complexity, amount and type of motion, etc). And these algorithms would need to change encoding parameters virtually instantly, as total encoding delay is typically about one GOP.

Another interesting discovery was an explanation of the AVC deblocking
filter. Seems like a completely separate algorithm within AVC, which can
just as easily be applied to MPEG-2, or any other block-based algorithm,
as it can be to AVC. (Very likely what Algolith did.)

No, deblocking CANNOT be applied to MPEG-2 per se'.

This would require replacement of hundreds of millions of DECODERS that do not currently have this feature. You are correct that any new product could add post filtering to clean up the problems with an MPEG-2 stream, and there are a few products that do this. Unfortunately, most of the display processors in digital TVs make things worse, not better.

I might add that I discussed this possibility in my initial reply to your post about the Algolith product. These techniques are feasible in new products, but you cannot do anything about existing products, except to replace them. And if you are going to replace them, it makes far more sense to upgrade the decoder to be interoperable with both h.262 and h.264.

Another interesting and not-too-recent improvement was this Digigami


from end of 2005. Which claims that 720p can be compressed in MPEG-2
down to 3 - 7 Mb/s. I assume this means 720 at 24p, presumably DVD

Did you notice the caveat?

"It amuses us that our MPEG-1 VBR encoder can also match and outperform H.264 on many progressive encoding tasks at HD frame sizes."

There are two factors at play here:

1. All of the MPEG algorithms work better with progressive frame-based source. The addition of tools for encoding interlace to h.262 and now h.264 is an abomination that flies directly in the face of everything we know about entropy coding. Eliminate this requirement and you eliminate more than half of the IP in both standards.

2. HD source is much easier to compress than SD source - it has less entropy because it is sampled with far greater accuracy.

That being said, you cannot get something for nothing. Most of these demos are interesting but not very practical. As you say, what happens in the real world is what matters.

it's all about working within the sweet spot for any algorithm. Unfortunately, the only sweet spot that most program distributors care about is their bottom 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: