[opendtv] MTF and oversampling [was euro ivp]
- From: prinyar.boon@xxxxxxxxxxxxxxx
- To: opendtv@xxxxxxxxxxxxx
- Date: Tue, 14 Sep 2004 13:28:24 +0100
Speaking of oversampling:
At IBC the Northlight 2 film scanner boasts a 9k CCD line array which is then
downsampled to give a 4K H output.
The new Arri film scanner has a 3K x 2K 2D array - the mechanics of which is
sufficiently finely machined to enable them to perform a 2 pass sampling
operation with a spatial offset on the second pass for 6K operation.
You may or may not be aware that we [S&W] were chosen to provide reference
downconversion from 4K to 2K for the DCI StEM mini movie - file based
conversion. The depth of modulation on the 65mm material is simply stunning.
The StEM material was originally scanned at 6K x 4K and downsampled to 4K x 2K
in the scanner.
Regds
Prin
____________________Reply Separator____________________
Subject: [opendtv] Re: Europe now debates i vs p
Author: opendtv@xxxxxxxxxxxxx
Date: 14/09/2004 13:40
At 10:32 AM -0500 9/13/04, Doug McDonald wrote:
>Are you sure that is true for 720p cameras? If is it, then there
>is room for quite a bit ... a very lot indeed ... of additional
>detail in 720p pictures.
I'm not absolutely certain, but I'm fairly sure I'm correct.
The camera that is most likely the one you were watching is the
Thomson/Grass Valley LDK 6000 MKII, which currently dominates
installations in HD remote broadcast truck, due to it's ability to
switch between 720P and 1080i modes. The spec sheet for this camera
does not provide any details about the frequency response. There is
an interesting spec: 55% MTF for 720P at 27 MHz.
There are several issues here.
First is the ability of any given sensor to acquire high frequency
details. Obviously, a 2 Mpixel sensor can capture higher frequency
detail than a 1 Mpixel sensor. But the difference may not be easy to
perceive because of the MTF issue. A slightly lower resolution
picture may look sharper if the MTF of the higher resolution camera
is lower. In essence, the high frequency details lack contrast when
the MTF of the camera is reduced.
Mark Schubin may correct me if I am wrong, but the 55% MTF at 27 MHz
suggests that a fine black detail at 27MHz will be captured as a
midtone grey. The lack of contrast will make it difficult to perceive
the detail. For example a B& W herringbone pattern may turn into a
solid grey pattern.
In order to capture the high frequency details with GOOD MTF, you
need to oversample, which is another benefit of the Thomson LDK 6000
(although most of the oversampling is in the vertical, not the
horizontal). For a 720P capture the LDK 6000 does some horizontal
oversampling - 2048 samples are resampled to 1280. This is not quite
the 2:1 oversampling that would be desirable, but it is better than a
1920 x 1080 sensor being used for 1920 x 1080 with NO oversampling.
Another issue is the overall response of the production chain - what
happens as the HD signals move through the production switcher,
signal distribution, and finally MPEG encoding.
Bottom line, you are correct. There is still plenty of headroom for
improvement.
>Actually, if I had not been careful to avoid it, it could
>also have been real content from computer generated
>stuff. The first day our local NBC station did HD, they
>had a computer generated bug that had very high frequency
>content indeed.
Yes, it is easier to generate high frequency details in graphics than
it is to capture equivalent levels of detail with a camera. But there
are some trade-offs here too. The higher the frequency of the graphic
edges, the greater the problems with MPEG compression, which tries to
quantize away the high frequency coefficients. It is better to
roll-off the response on the graphics a bit to help the downstream
encoder.
Remember, hardly anyone has a display capable of resolving these high
frequency details, so why stress the system? A spectrum analyzer lets
you "see" stuff that is not perceptible on the display.
>
>If indeed there is nothing real above 25 MHZ, then there is
>no need for 720p horizontal resolution. For a couple of
>weeks before yesterday Fox broadcast 720p football signals which
>were camera generated at 720p, cut down to professional grade
>digital, not MPEG, 480i, and then upconverted to 720p. Those
>signals did indeed cut off at about 20-22 MHZ. And they looked
>good ... real real good indeed ... but not the wow factor of the
>real 720P.
720P (and all the other formats for that matter) are just containers
for samples. How good those samples are is the real question. The
same could be said for NTSC: it took about four decades to max out
the NTSC container.
The ability to fill the spectra for any HD format will improve over
time, as we are able to deal with the harsh realities of high
resolution image capture. We need to oversample images to fill the
spectra; unfortunately, this means that we need to divide the
available photons among even more sensor sites. Its all about the
signal to noise ratio and sensitivity. The solution is likely to be
found in a new generation of sensors that do not have the physical
limitations of the venerable old CCD. keep an eye on CMOS sensors.
>
>If they are not getting any real signal above 25 MHz, they
>need to get better cameras, considering the alias point is 37 MHZ.
All things in good time.
For now, I believe that 720P is the HD sweet spot.
- It has more than adequate resolution for virtually ALL consumer
display applications. And it is now possible to build affordable 1
Mpixel LCD, DLP and LCoS display chips.
- It is possible to do some oversampling during 720P acquisition,
which improves the MTF.
- And 720P is a far better optimized for MPEG-2 compression that 1080i.
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.
This email and any attachments is confidential, may be legally privileged and
is intended for the use of the addressee only. If you are not the intended
recipient, please note that any use, disclosure, printing or copying of this
email is strictly prohibited and may be unlawful. If received in error, please
delete this email and any attachments and confirm this to the sender.
----------------------------------------------------------------------
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:
- » [opendtv] MTF and oversampling [was euro ivp]
