Browse opendtv: This Month's Archive • Main Archive Page • Related posts • Previous by Date • Next by Date

[opendtv] Re: HD standards conversion

From: Craig Birkmaier <craig@xxxxxxxxx>
To: opendtv@xxxxxxxxxxxxx
Date: Thu, 4 Aug 2005 07:47:24 -0400

At 7:13 PM -0400 8/3/05, Manfredi, Albert E wrote:
>For example, from the original MPEG stream, using
>the original interframes and motion vectors,
>recompute interframes to fall correctly on their
>new time slots, and scale the B and P frames for
>the new frame rate as well.
>
>Sounds CPU intensive, but for non-real-time
>conversion this might be doable.

Afraid not Bert.

The motion vectors in MPEG are not based on tracking the objects in 
the original source. They are a crude attempt to find block matches 
between frames and in some cases may actually track the actual 
motion. But in some cases they will just find an unrelated block that 
matches and use it. But the real problem comes from the stuff that is 
missing.

MPEG relies upon something very important to work. The original 
information. After all of the block matching, motion vectors and 
other tricks to create a prediction, that prediction is subtracted 
from the original frame and the differences are encoded using the DCT 
transform.

So even IF you could use the block and motion vector information to 
make an in-between prediction for a new frame, what would you 
subtract it from?

As compression technology improves it is likely that we will move 
toward REAL motion compensated prediction of the scene content as 
opposed to crude block matching techniques, and that doing real frame 
rate conversions will be feasible. But we are still a long way from 
that point.

To compound the problem, Bert might have noted that Mark was asking 
about converting from 1080@25i to 1080@ 29.97i.  Interlace makes all 
of this even more difficult, as the source is sub sampled and much of 
the needed information is missing.

If you want good motion compensatred prediction the place to start is 
getting rid of interlace. The next step is temporal oversampling so 
that you have more detail (about the objects) to work with. You can 
do some really nice stuff by acquiring images as 1280 x 720 @72P, 
even if the content will ultimately be released as 24P.

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.

References:

[opendtv] Re: HD standards conversion
From: Manfredi, Albert E

[opendtv] Re: HD standards conversion

Other related posts: