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

  • From: Ron Economos <w6rz@xxxxxxxxxxx>
  • To: opendtv@xxxxxxxxxxxxx
  • Date: Fri, 20 Jun 2008 01:56:55 -0700

The hot product that re-encodes from baseband is the Motorola
DSR-6050 IRD.

http://www.multichannel.com/article/CA6480828.html

The box contains an MPEG-4 (AVC) IRD glued (over baseband)
to an HD MPEG-2 encoder and allows for MPEG-4 (AVC) satellite
feeds to be used at MPEG-2 only head-ends (which is pretty much all
of them). The HD MPEG-2 encoder is from Magnum Semi (formerly
LSI Logic (formerly C-Cube)). In fact, I have one in the lab
right now.

BTW, the 8x8 transform (and variable quantization matrix)
was put back into H.264 High Profile after it was found that
the 4x4 transform looks like shit at HD resolutions. Basically the
4x4 transform (with it's fixed quantization matrix) cannot always
generate enough coefficients to provide a sharp image. Today,
almost all broadcast HD H.264 is using High Profile.

Ron

Craig Birkmaier wrote:

At 10:51 PM -0700 6/18/08, John Willkie wrote:

yeah, the box, fool, is practically useless for that, since, uh, NOBODY takes a compressed MPEG-2 stream back to baseband, then uses this filter, then recompresses it. And a "prefilter" would imply that it goes before any compression, not in the middle of a decompression/compression daisy chain. Kind of hard at the beginning of a chain to know the noise that would be encountered in the middle of the chain.

Indeed, taking a compressed MPEG-2 stream down to baseband would actually impose noise JUST BY ITSELF since MPEG-2 compression is lossy and asymmetrical. So, you'd be lucky if the "prefilter" was able to remove the noise caused by the decompression to baseband.

The market for this device, Bert, is most likely YOU. Let us know how you fare with it.


Actually John, there are many situations in which a broadcaster or cable head end might be trying to clean up a compressed feed then re-encoding it for emission.

Many cable systems utilize the HITS (Headend In The Sky) service.

http://www.hits.com/

This service (I believe it is operated by Comcast) compresses a wide range of cable networks and distributes them to cable head ends via satellite.

In most cases these feeds are not decompressed, however they may be groomed by devices like the Cherry Picker to cram more stuff into a cable multiplex.

Because of the high cost of satellite bandwidth, many syndicated programs are now distributed to broadcast facilities via satellite as compressed MPEG-2 bitstreams. And then there is the age old practice of capturing the broadcast signals from other stations in a market to source clips for use in newscasts (with permission of course!). So it is becoming far more common for the feeds coming into a station to arrive as compressed MPEG-2 bitstreams.

The MNR mode of the Algolith device is specifically designed to deal with quantization noise, also know as mosquito noise. This type of distortion is much more predictable than the random/Gaussian noise found in uncompressed sources. It typically occurs at the boundaries of high frequency details, so a clever algorithm can look for these details then process the distortions around them.

You are correct that MPEG-2 creates "noise," but the noise is correlated to image detail, not random, and thus may be reversed to some extent if the damage is not too severe.

And yes there may be a consumer market for devices that undo the damage of video compression. Some of this was incorporated into h.264 decoders (the deblocking filter tools). And h.264 dropped the 8 x 8 DCT quantizer entirel, replacing it with a scalar quantized that is less complex and more accurate:


http://ieeexplore.ieee.org/Xplore/login.jsp?url=/iel5/76/27384/01218192.pdf?temp=x

Summary: This paper presents an overview of the transform and quantization designs in H.264. Unlike the popular 8/spl times/8 discrete cosine transform used in previous standards, the 4/spl times/4 transforms in H.264 can be computed exactly in integer arithmetic, thus avoiding inverse transform mismatch problems. The new transforms can also be computed without multiplications, just additions and shifts, in 16-bit arithmetic, thus minimizing computational complexity, especially for low-end processors. By using short tables, the new quantization formulas use multiplications but avoid divisions

So h.264 provides a significant improvement in arithmetic accuracy and the visibility of artifacts is reduced by acting upon smaller 4 x 4 regions of the image.

So my educated guess is that products like the Algolith device will not find their way into TV receivers. The shift to improved compression techniques will make this unnecessary. And we can hope that broadcasters and other program distributors have learned that excessive compression is

BAD!

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: