[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Information, Compression, and storage.

Let's start with a little thought experiment:

Suppose we have half an hour's worth of film, and some sort of bit bucket,
be it a hard disk array or digital tape format, that can store that half
hour at 720 pixels by 480 pixels, uncompressed.  By applying 4:1 compression,
we could store the same show in the same bit bucket at 1440 x 960 pixels.
Digital beta uses about 3:1, and HDcam 7:1 compression, and they look very
good.  So, our 4:1 compressed show would certainly look better than the
lower resolution uncompressed show.  But if we had gone to 10,000:1 compression,
even though we 'd have 72,000 x 48,000 pixels, the compression artifacts would
be severe, and the show would look like absolute crap.

The point is, like all the other variables you adjust as colorists, the
right place is never at one extreme of the range or the other.  Selecting
where in the range that right place is, is why you make the big bucks.  The
same thing will happen with compression.


I understand that with things like
>computer programs there are such things as Lossless compression (e.g.
>PkZip and PkUnZip -

Lossless compression is a very different thing than the DCT based MPEG
and JPEG that we use for pictures.  In general, it's very difficult to
get useful ratios -- greater than 2:1 -- from lossless schemes.  Also,
the actual ratio from a lossless scheme is highly variable and depends
on the content of the data.  There is a deep and interesting theorem in
the mathematics of this that states that for any lossless compression
scheme, there exists a data set such that the supposedly compressed
output is larger than the input data.  Zip and the like are practical
because your computer files generally don't look like that pathological
data set, and the rare ones that do are more than offset by the ones that
compress nicely.

>From what I understand NTSC Broadcast is 3.5 Megabytes per second ( 300
>KB per frame),  when you talk about digitized signal for broadcast. Is
>there a way to correlate that to Analog?

NTSC is purely and completely analog.  There is no way to give a hard and
fast equivalent in digital terms.  But NTSC is quite highly compressed in
its own analog way.  First, it uses interlace, which is a sort of ham-
fisted lossy compression, throwing away half the lines.  Then the color
is sampled at half the frequency used for luminance, because the human
visual system doesn't resolwe color as much as brightness.  This was
carried forward into digital, which is why we have 4:2:2 and all that.
Third, to make room for the chroma information, fine horizontal detail
from 3.58 MHz and up, was thrown away.  Digital does a much better
job of compression because it's better at throwing away things that
people don't notice. Mixing analog and digital compression techniques
is a mistake, because they're both trying to get rid of the same kind
of redundancies, and digital does it better.  That's why interlace has
no place in any digital TV system.

>Basically I'm curious about, NTSC Broadcast Information rates, compared
>to: Standard NTSC Transfers,  1K transfers,  2K transfers, Data Rate
>Transfers, HD transfers (uncompressed if there is such a thing?). Also
>I'm curious about how much these figures change once you add in

Both HDTV and NTSC use a 6 MHz channel.  With the (controversial) 8VSB
coding chosen by the ATSC, that channel has a capacity of 19.4 Mega bits per
second.  Uncompressed HDTV can in fact be recorded on the D-6 format, which
is sort of like D-1 only more so, and on the obsolete open reel one inch
Sony HDD-1000 machines.  (Sort of a 1"C on steroids)  The data rate for
uncompressed HDTV is 1.485 Giga bits per second, which means that the
compression ratio for ATSC HDTV is about 76.5:1.  That's on the ragged edge
of what's acceptable for viewing.  The idea was that compression technology
would get better, and the pictures would get better using the same decoders.

>I don't understand how much compression 50:1 actually is,

Nobody does.  50:1 could look nearly lossless, or it could look
like crap, depending on how it's done, and what the images are.

>I am also interested, because I am seeing really bad digitizing
>Artifacts popping up on Television. Most recently watching a program on
>one of the BIG three Networks, and their pre-taped introspective pieces
>all had really bad compression. Jaggy movement, and skin tone gradations
>that made people look like they were part of a topographical map.

Was this on their regular NTSC service or on the new digital channels?
I've seen stuff work and fail on the digital channels, and transmission
interference and weak signals all can make artifacts that look very much
the same as compression artifacts.  Error correction or concealment are
part of the compression schemes, and when you get more errors than they
can handle, the way it breaks down looks the same as bad compression.

If it was on NTSC, either somebody wanted it that way, or they used
something cheap along the way (maybe NLE?), and decided to tolerate it.

The other thing to watch out for is that amounts of compression that
work fine for just recording and playing something back can be too much
for use with other post production processes, such as tape to tape
color correction, re-sizing, or compositing.  Mixing different kinds of
compression can also be a problem.

-- J.S.

Thanks to Rich Torpey for support in 1999
No advertising/marketing allowed on the main TIG.  Contact rob at alegria.com
anonymous messaging now at http://www.alegria.com/HyperNews/get/ubique.html
1045 subscribers in 41 countries on Tue Jul  6 14:01:41 CDT 1999 
subscribe/unsubscribe with that Subject: to telecine-request at alegria.com
complete information on the TIG website http://www.alegria.com/tig3/