>"Gary Pollard"
<gpollard@allivigator.com> wrote in message
>news:b7tqve$kas9@imsp212.netvigator.com...
>> No it doesn't only exist with DV or even only with
digital formats. I've
>> been in broadcasting for 20 + years, much of which
we've used analogue
>> formats, and we have ALWAYS asked guests not to
wear checks or too fine
>> pin stripes for precisely this reason. Even if we
are recording directly
>> onto 1" or 2" analogue tape.
>> Gary
On Sun, 20 Apr 2003 22:39:18 +1000, "Native_MetaL"
<NativeMetaL@optusnet.com.au> wrote:
>ok thanks Gary do you know whether there is any kind of
lens filter that can
>be used to reduce the problem?
For a wonderful, REALLY complete discussion of this,
see John Dyson's post on this (unfortunately detached from
this thread...), below. The short of it: to reduce
artifacting,
either allow no motion in subject or camera; reduce subject
and lighting contrast as much as possible; add a filter to
the lens that reduces detail considerably; or accept the
aliasing (which does vary from camera model to model, and
among formats). (I prefer to choose a suitable camera,
and then accept the consequences...;-)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The moire or the more general problem called 'aliasing' can
occur
on any sampled video format. However, when the sampling and
compression are properly implemented, the 'aliasing' will be
well controlled.
Here are some facts about the problem (in no particular
order):
1) Moire is one
manifestation of a more general problem called
'aliasing.'
A) Aliasing
happens in a sampled system, where the signal has
higher frequency
components than what can be sampled and
reconstructed. As a matter of
mathematics, the sample frequency
has to be twice
the maximum frequency of the signal that is being
sampled. The signal might even have noise components
that are
outside of the
desired signal frequency range, and the noise that
is higher than
the range that can be sampled and reconstructed by
the sampling
process will cause aliasing components.
2) Moire (or
aliasing) can even be manifest on FM video recorders
like VHS, SVHS,
BetaMax, BetaCam SP or even broadcast 1".
Yes,
ANALOG formats
can produce aliasing because the FM sidebands as
recorded on tape
can wrap around 0Hz (DC.)
3) The
pre-filtering needed to minimize the aliasing problem can
decrease the
apparent sharpness and apparent resolution.
4) Sometimes, on
non-critical applications, the amount of pre-filtering
can be minimized
to help compensate for potentially low resolution
or detail from a
lower quality source. The apparent
increase in
detail will be
bought at the cost of more aliasing artifacts.
5) The moire and
other aliasing artifacts will create a permanent
imprint on the
video signal. Once aliasing appears,
then removing
it will be much
more difficult than removing the aliasing before
the sampling
process. As a matter of physics and
mathematics,
removal of
aliasing after it has appeared in the signal is impossible
in the general
case, but sometimes the signal can be cleaned up a
little.
6) In sampled
picture element TV cameras like CCD units, to avoid
aliasing, the
image on the CCD should be low pass filtered (optically
remove the high
frequency components that cannot be properly sampled
by the CCD
array.) On older tube based cameras,
even though the
horizontal scan
was 'continuous', the scanlines essentially sampled
(and still does)
the video in the vertical axis.
A) An especially ugly form of vertical aliasing
is manifest by
'interlace
twitter.' Even though 'interlaced'
video definitely
needs some
kind of pre-filtering to mitigate the twitter, progressive
video can be ugly without pre-filtering the
video also. When
trying to
reproduce too much detail, even with progressive video,
there are
ugly stairstepping artifacts and even the possibility
of larger
area 'beat products' that appear as moire.
B) Even for the old analog TV cameras, the
scanlines represent
a sampling
in the vertical dimension, and so sampling issues
had to be
dealt with for highest quality.
C) For alias free reconstruction, CCD cameras
need an optical
filter that
removes the high frequency detail that cannot
be properly
sampled by the CCD imaging array. If
the CCD
array has
too few pixels, sharpness can be improved by decreasing
the amount
of optical prefiltering, but aliasing is made
worse. With CCDs, it isn't possible to change the
pixel
size to
overlap between adjacent samples to help to filter
out the
excess detail. With the old common TV
camera tubes,
the spot
size could be made large enough to allow the overlap.
D) One approach used in the old tube TV cameras
was to make the
scanning
spot larger than a scanline so that excess vertical
detail will
be removed before sampling. The problem
with
increasing
the spot size was that it would often grow the
spot in both
the vertical (desired) and the horizontal (less
desired)
directions. The aperture correction
helps to compensate
the video in
the horizontal direction for the large (larger than
one
scanline) spot size that mitigates normal spatial aliasing,
and the
uglier time dependent aliasing from interlace.
E) Some very extreme conditions of aliasing
like behavior, (e.g. in
the case of
FM wrap-around on analog VTRs) can look like
large area
waves moving through the video.
F) Quantization and truncation of DCT
coefficients (one of the steps
in MPEG and
DV type video compression) can appear to be similar to
the optical
aliasing effects. Removing or changing
the coefficients
can create
nonlinearities that effectively add distortion signals
that are
high frequency, and can appear similar to too much detail.
I haven't
seen 'moire' from the errors caused from compression
schemes that
remove redundancy by using DCTs, but I have definitely
seen effects
that look like the stair-step form of aliasing.
G) Even the DV25 compression artifacts appear
to produce low frequency
components
that are likely due to high frequencies caused by the
nonlinearity
and 'beating' with the sample frequency (result: aliasing.)
This means
that the 'low frequency' components become a permanent
attribute of
the video signal, and unfortunately, VHS decks reproduce
the low
frequency aliasing very well. In
essence, DV25 (normal DV)
can produce
artifacts that can even degrade the image quality of
a VHS (yes,
just VHS) distribution copy.
H) Digital recording schemes that have less
compression tend to produce
fewer extra
high frequencies (because they have to truncate less
detail), and
so the excess aliasing from the compression record and
playback is
nil.
I) Very high quality TV cameras have more CCD
imager elements (sometimes
both in the
H and V directions), and allow for less optical filtering,
and can
support the more precise electronic digital filtering. The
optical
filter still has to remove the detail that cannot properly be
sampled by
the CCD imager, but CCDs with lots more elements can allow
for less of
a sloppy optical filter. With less
optical filtering,
then need
for 'compensation' for the optical filter (which will obviously
remove some
desired detail also, alot like the spot size on a tube
TV camera)
by boosting/compensating the VALID high frequencies
will be
lessened. This will allow for both less
noise and less aliasing.
J) Even HDTV 720p (progressive) needs filtering
to avoid the ugly picture
from
stairstepping and moire. Interlaced
video is more critical, because
the aliasing
that jumps around is not just ugly, but it is distracting.
Interlace
filtering isn't as bad as it initially sounds, however, because
the
electronic portion of the filtering can be made dynamic. It is
very
difficult to create a dynamic 'optical' filter, but an electronic
filter,
especially when using a CCD array that has LOTS more picture
elements
(and perhaps like mine, has double the scanlines), can allow
for better
detail than interlace normally would have, yet also remove
by far most
of the aliasing artifacts.
I hope that this gives some help in understanding the
effects of aliasing
for simple video applications. Where aliasing in video can cause
moire, stairstepping and, in certain cases, some dynamic
artifacts that
mix aliasing with interlace. Equvalent kinds of aliasing in audio
applications can cause wierd tweets when an instrument (or
multiple
instruments) mix with the sampling frequency. In cases where the
aliasing effects are extremely obvious, similar to when the
stairstepping
can happen in video, with audio, the sound can be 'harsh',
even if there
are no extremely gross effects. It is VERY IMPORTANT to remove the
out-of-band signal components (even if noise or distortion)
before
the sampling mechanism is applied to the signal. One not-so-obvious
case for audio, where out-of-band artifacts should be
removed, is to
properly dither the audio before sampling. A low level dither
(properly shaped) will help to mitigate regular patterns of
aliasing
artifacts, and spread them out to be less perceptable.
(This was written without a careful review -- it is too
late, so
I hope that I am not misleading anyone, and hope that this
is adequately
complete.)
John
Dyson
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~