[opendtv] Re: Why Europe should choose 720P for HDTV
- From: Craig Birkmaier <craig@xxxxxxxxx>
- To: opendtv@xxxxxxxxxxxxx
- Date: Mon, 13 Dec 2004 09:51:33 -0500
At 11:52 AM -0600 12/12/04, Doug McDonald wrote:
>
>Uh, no on that last one. I am seeing blurring that occurs inside
>the camera. For example, a white line marker in football retains
>sharp edges with a horizontal pan, and everything else shows
>the same blurring. Also ... the blurred frames should be
>much easier to compress since the originals lack
>much detail at high horizontal frequencies.
Wrong on both counts.
IF the lines are sharp, but the objects moving over them are not,
then it is NOT the camera. What is happening is that it is easier for
the MPEG-2 algorithms to match blocks where the lines are seen, than
it is for the algorithm to deal with random motion. Nothing new here.
If you are saying that the lines and the objects that are moving both
exhibit the same blurring, then you ARE describing the filtering in
the camera.
As someone pointed out, the actual amount of blurring is a function
of the exposure time for the frame. Higher shutter rates provide
sharper edges on the moving objects, but may lead to discontinuities
in motion if the frames are too sharp.
It is also a fallacy to believe that softer images are easier to
compress with MPEG-2. This may seem a logical conclusion, since low
pass filters are used to reduce stress on encoders, but this is just
a band-aid for a larger underlying problem.
MPEG-2 encoders can only work as well as the block matching
algorithms work. If you can get GOOD predictions (matches), you can
make the encoder work MORE efficiently. If you can;t get good
matches, then you start to see encoding artefacts. Eliminating the
blur on the edges of moving objects can HELPthe block matching
algorithms in an MPEG-2 encoder because it is easier to track the
motion precisely. Gary Demos did a number of experiments with this,
working with the Polaroid 720P camera that was modified to shoot at
72P. he also did similar tests with film source shot at various high
frame rates with different shutter angles (i.e. more or less blur on
the moving objects.
What Gary found was that at higher frame rates, with less blur, the
MPEG-2 tools worked better, allowing for longer GOP sizes that
improve encoding efficiency. This is logical, since we are dealing
with an entropy coder. With MORE samples that are MORE accurate, the
crude MPEG-2 block matching tools work better.
But there are limits to this. Most notably the channel bandpass,
which places a fixed ceiling on the output bitrate of the encoder. As
I have mentioned many times before, the 19.3 Mbps bandpass of the
ATSC channel is marginal for difficult source material. And when you
concatenate codecs the situation only gets worse.
And if you really want to stress out, consider what happens with
1080@60i for sports. Now you are blurring more to allow for the use
of interlace, and half of your images are missing, creating all kinds
of gaps in the spectra which increase the entropy the encoder must
deal with. But you are still trying to push as many pixels through
the encoder as you are with 720@60P. 720P will win this battle every
time.
I strongly suspect that Fox has tested their entire system from truck
to display to determine the best way to use their 720P cameras. They
may have run test to determine which shutter speed is best to deliver
adequate sharpness in moving objects without stressing the encoders.
But there may be more involved in this decision than meets the eye.
The 720P source must also be downconverted for the NTSC channel. It
could be that the NTSC pictures look better when there is more motion
blur (i.e. a lower shutter speed).
At any rate, this post should point out that there are many variables
that aaffect the quality of the delivered HDTV imagery.
Regards
Craig
>
>The killer argument is that the same blurring is seen
>on the analog show, which never had MPEG-2 done to it.
>
>Doug McDonald
>
>
>----------------------------------------------------------------------
>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.
----------------------------------------------------------------------
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: