[opendtv] Re: Image quality
- From: Craig Birkmaier <craig@xxxxxxxxx>
- To: opendtv@xxxxxxxxxxxxx
- Date: Fri, 11 Nov 2005 09:47:11 -0500
At 10:22 AM -0800 11/10/05, dan.grimes@xxxxxxxx wrote:
>"We do a lot of dancing around about 1920 x 1080 when it's becoming less
>and less likely (with smaller and smaller imagers) that there will be any
>significant energy at the high end of that."--Mark Schubin
>
>So is there any benefit to produce in 1080?
It is not the number of samples per line or number of lines that are
important per se'.
What is important is oversampling relative to the format that you
plan to use for emission, and care to make certain that the integrity
of the samples that will be encoded for emission are of the highest
quality. Garbage in = garbage out.
The reason that ABC is using 1920 x 1080@24P for production then 1280
x 720 @24p for emission is that this is the proper way to do things.
It allows the production system access to the highest quality
samples, which can be very important for special effects processing.
The process of resampling eliminates entropy, which is the biggest
enemy of digital compression. When I say entropy, I am talking about
sampling errors - i.e. sample values that are not representative of
the original source. When you re-sample to a lower resolution the
filters take information from a range of surrounding samples to form
the new sample - the net result is that the new samples are more
accurate, and more picture detail is available under the MTF curve.
That being said, there is a world of difference between 35mm film and
a 1080i HDTV acquisition system. The film source is oversampled, and
the electronic scanning process is able to sample the film source
with excellent MTF, preserving much of whatever high frequency detail
that may exist.
On the other hand, most 1080i acquisition system are either
undersampled, or the sampling is one-to-one for the target raster.
The cheap 1/3" cameras are undersampled - the sensors typically have
no more than 1440 samples per line and possibly even less. The
acquired samples are then resampled to about 960 samples per line for
compression and recording. And with HDV, the integrity of the sample
may be impacted by the high levels of quantization that are necessary.
For the high end 1080i cameras, most of the sensors are 1920 x 1080 -
some of the Thomson/GVG cameras have 2048 samples per line, but this
is not sufficient oversampling to contribute much to better MTF. To
do 1920 x 1080 right you need to sample at AT LEAST 2880 x 1620 with
progressive frames, then resample to either 1080i or P. 2x
oversampling is even better.
Since most 1080 line cameras do not oversample, and they use
interlace for high temporal rate acquisition, the result is that the
potential spectra of the format is not filled. Another way of saying
this is that the frequency response for 1920 x 1080 should extend out
to at least 27 MHz. Unfortunately, when you sample at the same
density as the format you are trying to create, most of the
information above 20 MHz is just noise, or has virtually zero MTF -
i.e. NO DETAIL.
So the bottom line is that it is almost always useful to produce
using a format that is of high quality than what you plan to use for
emission. But it is not necessarily true that a camera that claims to
use a higher quality format will produce better pictures than one
with a lower density raster that is properly oversampled.
Today many professionals are shooting with high quality SD cameras
that use considerable oversampling in the horizontal axis, and
progressive sensors. For example, the Panasonic SDX-900 uses 2/3 inch
progressive sensors that have 520,000 samples (approximately 1080 x
480). With the larger 2/3 inch sensors and the oversampling, the
camera produces very high quality 24p images that many people are
bumping up to 720P.
>Our new facility will probably be 1080@30i. Not because I chose it but
>because I want to be compatible with out PBS station for many reasons (they
>do our master control for us). But it also seems like a lot of equipment
>now days will allow one to produce in any format one wants when they want.
>True?
Corect. So don;t waste your time trying to be compatible with the
formats being used by your PBS station. Resampling to 1080i for
emission is not difficult. You may be better off shooting 720P and
waiting for the next generation of low cost HD camcorders that will
not be so severely constrained by the use of MPEG-2 long gop
compression (HDV). Personally I would wait for the new Panasonic P2
camcorder that will shoot both SD and HD at multiple frame rates.
This camera will use the DVCPro-HD codec which is frame based at 100
Mbps. And I would not be surprised if Panasonic adds H.264
compression to some products by next year.
>
>But regarding the lens, I have seen demonstrations of 1080 and 720 with the
>same lenses (leica primes and angineau zoom) and found 1080 to look better.
>I realize that this is with some of the best lenses and top quality
>cameras. I can understand how there is no way to get the resolution
>necessary from a lens that shrinks the image down to those tiny imagers.
>So in reality, if one is using industrial grade HD equipment, I take it to
>be a case of "why bother"?
If the 1080 camera used a 1920 x 1080 sensor and the 720 camera used
a 1280 x 720 sensor, then I would expect the 1080 line camera to look
better. But if the 720 camera uses a 1920 x 1080 sensor (1.5:1
oversampling), i would expect it to equal the performance of the 1080
line camera and produce source with better MTF.
And Mark nailed it with respect to the limitations of 1/3 inch sensors.
HDV is a low cost stop gap format. Better things are just around the corner.
>But then, I have often seen cases where oversampling helps, even when the
>apparent high frequency energy is supposedly not there. I have seen it in
>audio, and certainly in photography. Is there no correlation?
I talked about the correlation above. Oversampling reduces sampling
errors at ALL frequencies. And it improves MTF at the higher end of
the curve, where fine details are filtered into grey oblivion.
As Mark will tell you, the human visual system is more sensitive to
contrast than to fine detail.
If you shoot B&W line pairs - think multiburst - the contrast will
decrease as the frequency increases. At some point those B&W lines
will become a grey blur. The useful information - to the human visual
system - is the stuff with good contrast. This is why a good SD image
can look as good as a poor HD image, and you can upsample a good SD
image and it will still look good on an HD display.
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:
