NetNews Usenet Archive 1992 #16/NN_1992_16.iso/spool/comp/compress/2857

Internet Message Format | 1992-07-28 | 3.9 KB

Xref: sparky comp.compression:2857 rec.video.production:1187 rec.video:9719
Path: sparky!uunet!zephyr.ens.tek.com!tvnews!taupe.tv.tek.com!dmc
From: [e-mail] (Donald M. Craig)
Newsgroups: comp.compression,rec.video.production,rec.video
Subject: Re: Compression artifacts in NBC Olympic coverage
Keywords: Standards Conversion, Olympics
Message-ID: <[e-mail]>
Date: 29 Jul 92 01:39:55 GMT
References: <[e-mail]> <[e-mail]> <[e-mail]>
Sender: [e-mail] (news user)
Reply-To: [e-mail]
Followup-To: rec.video.production
Organization: Tektronix TV Products
Lines: 59

We've been monitoring the Olympic feeds here in our lab, and
the artifacts being discussed are the results of PAL->NTSC
standards conversion.  The cameras at many venues are indeed
using fast shutter speeds, but that only reduces the amount
of blurring in each individual field of video.  We have seen
two kinds of standards conversion algorithms, the expensive
and the inexpensive.

PAL, the European television standard, has a 50Hz. field (picture)
rate, while NTSC, the North American standard, has a 60Hz rate.
There are 5 PAL video fields (pictures) for every 6 NTSC ones,
and the problem faced by a PAL->NTSC converter is to create
a new NTSC field that lies between a pair of existing PAL ones.
After an NTSC field and a PAL field line up, the next NTSC
field is 5/6ths of the way towards the next PAL field.  The
NTSC one after that is 4/6ths of the way towards the next PAL
field, the one after that 3/6ths of the way, the one after that
2/6ths of the way, the one after that 1/6th of the way,
and the one after that lines up with the corresponding PAL
field again.  An expensive standards converter performs motion
estimation on small blocks of pixels from one PAL field to the next.
It then assumes linear motion from one field to the next,
interpolates e.g., 5/6ths of the way along for the first NTSC
field after a PAL/NTSC lineup, and generates the new NTSC field
pixels accordingly.  Subsequent fields would interpolate 4/6ths,
3/6ths, 2/6ths, and then 1/6th  of the way along each newly
calculated motion vector.  Each motion vector is calculated for
each small block of pixels in a field.  This process works quite
well, although accelerated motion won't be quite right, and that's
the major problem we are seeing.  For the coverage we analyzed
(women's diving) that used the expensive converter, it took 3
output fields for it to stabilize after a cut, during which time
the output was fiction.  It also made the most marvellous Leroy
Nieman (sp?) paintings out of the vertically panned backgrounds,
and for one glorious field chopped the hands from Inga Afonina's
arms and glued them to her knees. There was probably also
something wrong with the stupid thing, since linear horizontal
left to right wipes, as John Abt pointed out, acquired extra
`ghosts' on the right hand side, that disappeared every 6 fields.

An inexpensive standards converter doesn't do any motion estimation
on small blocks of pixels.  It doesn't do any motion analysis
at all.  It simply mixes two adjacent PAL fields together in
the ratios 17%/83% for the first NTSC field after a match,
33%/67% for the second, 50%/50% for the third, 67%/33% for
the fourth, and 83%/17% for the fifth.  So if, for example,
you watch a vertical pole against the sky while the camera pans
horizontally past it, every 6 NTSC fields you see one pole, and for
the 5 fields in between you see two poles, one getting lighter while
the other gets darker.  Some (but not all) of the gymnastics coverage
was uplinked with this type of conversion.

The above discussion omits the other PAL->NTSC conversion problem,
which is converting PAL's 625 line frames to NTSC's 525 lines.
That's a much easier problem to solve adequately.  It also slightly
simplifies the actual frame rate differences and omits entirely the
color encoding differences.

Don Craig    Tektronix Television Systems Group