Stephen Neal
Guest
|
 Posted:
Sat Oct 30, 2004 5:28 pm Post subject:
Re: Are there any good articles on HD --> HD conversion? It |
|
|
"HireMe.geek.nz" <mikehack@u.washington.edu> wrote in message
news:clvk6q$7o5$1@gnus01.u.washington.edu...
| Quote: | Are there any good articles on HD --> HD conversion?
|
Not sure - the Snell and Wilcox site is always worth a look - they are at
the bleeding edge of standards conversion techniques. (They created, and
successfully patented DEFT 3:2 pulldown removal techniques for 60i>48i>50i
conversions of 24p source material, and have some of the best 60i>50i and
50i>60i converters using their PhC Alchemist phase correlation based
converters)
| Quote: | It sounds easy, but is probably more difficult than PAL --> NTSC
|
Well PAL->NTSC is pretty much a legacy thing - certainly almost no
broadcasters use composite PAL or NTSC for production these days. Conversion
is pretty much universally from 576/50i to 480/60i digital components - with
no PAL or NTSC subcarriers involved. (There are specialised and very high
quality PAL and NTSC digital decoders available where PAL and NTSC sources
are still required to be decoded to digital component - say for archive
conversion of 2" or 1" VT from the 70s and 80s. This is also an issue in DVD
remastering - as DVDs are neither PAL nor NTSC but are also digital
component - as are SDTV digital TV systems)
Composite standards conversion is pretty much limited to news production
these days I guess - and even then many location trucks are now being based
around digital component rather than composite analogue systems - so even
then PAL/NTSC is less widespread than it once was. Increasingly DSNG trucks
are SDI over fibre, or MPEG2 digital radio, from the camera to the truck,
digital component vision mixers like Magic Daves are used in the trucks
(handily containing synchronisers on their inputs so reducing requirements
for genlock/timing) with the outputs of the trucks feeding MPEG2 encoders
for DSNG uplink with no PAL/NTSC in the way. This can mean a component
digital path from live camera to the home with no PAL or NTSC getting
anywhere near the pictures. Hurrah!
| Quote: | Certanly there are still 50 hz --> 60 hz issues, but of a more subtile
nature. Interlacing will be hellish for a long time to come, before all
its issues are resolved.
|
The issues with HD are mainly caused by the massively increased data rates,
and thus processing requirements, of the source and destination formats. I
don't know why you think the differences between 50 and 60Hz motion become
more subtle just cos the images are higher resolution spatially - the
temporal resolutions are identical to those of SDTV 50 and 60 systems -
there is no increase in the temporal (i.e. time sampling) resolution in HD -
and it is the temporal issues (i.e. the change in field rate which is the
temporal sampling rate) which require the clever processing whether it is SD
or HD spatially.
Modern converters have to recreate a 60 or 50 Hz image stream from a 50 or
60Hz source - they do this, ideally, by tracking all of the motion in a
scene and recreating new fields/frames in between the source fields/frames.
If you get this wrong you end up with blurred or jerky motion - as you will
if you use a cheap algorithm that doesn't try to track individual motions of
differing picture elements, but instead just blends fields in various
weights depending on the total amount of motion in a scene (as was state of
the art in the 80s). To get it right you have to have decent motion
tracking/detection algorithms - and there are various ways of doing this.
(The processing required for this is related to that required to MPEG2
encode stuff in some cases - as both conversion and compression require
accurate motion information)
In the case of HD - if you compare 720x576/50i (10368000 luminance samples
per second) source video with 1920x1080/50i (51840000 luminance samples per
second) source video then you have are dealing with at least 5 times the
amount of luminance samples in the same amount of time - more if you compare
720x480/60i (10368000) with 1920x1080/60i (62208000) which is 6:1. You
therefore need to have approx 5-6 times the amount of processing power to
implement the same algorithms that you'd use at SDTV resolutions. (You
could think about it by dividing the HD source picture into 6 tiles - say 3
x 2, each of which would be an SD resolution image - which you could then
feed to a separate standard definition converter - 6 tiles would be around
the SD resolution for 60i stuff - so it is the equivalent of requiring a
converter 6 times more powerful)
It may be that some elements of the converter don't need to be 6 times more
powerful - but the main data paths, video processing, storage etc. will need
to be this order of magnitude different AIUI.
IMHO the highest quality SD converters used for 480/60i<->576/50i conversion
using PhC style processing are still pretty state-of-the-art in processing
power terms - so scaling them up by a factor of 5 or 6 is still not quite
possible at a commercially acceptable cost - or is only just becoming so.
ISTR that NBC used a block-matching converter to convert from 1080/50i to
1080/60i for the US HD Olympics coverage (because the Olympic SD production
was 50i the HD really had to be 50i as well) and this, whilst good quality,
may not have been as capable as a phase correlation based system would have
been? There were observations of motion artefacts introduced by the
conversion I believe.
Steve |
|
FAQ
Memberlist
Register
Profile
Log in to check your private messages
Log in