[opendtv] Deinterlacing state of the art

From: "Manfredi, Albert E" <albert.e.manfredi@xxxxxxxxxx>
To: <opendtv@xxxxxxxxxxxxx>
Date: Mon, 7 May 2007 11:50:29 -0400
Ron Economos wrote:

> Just a thought. I wonder if the de-interlacing performance
> of the PDP and LCD panels was a factor in the lower 1080i@25
> scores. In the full report, the de-interlacing performance
> is characterized as "good". Was it good enough? Actual tests
> of panel de-interlacing can be pretty disappointing.
>
> Maybe displaying 1080i@25 on a good old truly interlaced CRT
> based studio monitor would have leveled things out?

This has to be a major factor to consider. Seems to me that
deinterlacing is one of those CPU intensive tasks that will improve in
time. I would predict that as deinterlacing algorithms improve, the
image quality comparisons will more closely track the theoretical limits
of baseband signal bandwidth or pixel transfer rate.

This article

http://www.digitaltvdesignline.com/howto/showArticle.jhtml?articleId=199
203870&pgno=1

describes a new technique that is supposedly an improvement of the
motion adaptive and motion compensated techniques.

The article says that motion compensation techniques measure the
reliability of their results. If the results measure unreliable, then
the algorithm defaults to spatial interpolation. And because these
algorithms are conservative, deinterlacing often will be done with
spatial interpolation, which will not give great results.

This new technique, which they call "super-resolution," looks at the
results of various deinterlacing techniques and selects the best one for
each occasion. They claim this:

"A criterion has to be found to choose one candidate estimator. This
criterion has to be comparative, using a mathematical norm on the video
signal. A total variation norm measures the amplitude of oscillations
and variations of the image (jaggies and flickers are spatio-temporal
oscillatory artifacts), without penalizing discontinuities. One can
prove that the total variation of an image is essentially proportional
to the total length of edges, whether these edges are discontinuous or
smooth transitions. This norm is then used to choose the appropriate
candidate."

The article shows an impressive result. Whether this is the holy grail
or not, it seems clear that Moore's Law will play a part in making 1080
at 60i a more competitive mode for HDTV, even if 1080 at 60p is not
incorporated in the standard or in production equipment real soon.

Bert
 
 
----------------------------------------------------------------------
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.
[opendtv] Deinterlacing state of the art

Other related posts:
