IP: The Appearance of TV Displays

[Dave Farber <farber () cis upenn edu>]

Bill is an EXPERT on this game djf


From: wfs () image mit edu (William Schreiber)


The Appearance of TV Displays


In the course of the Advanced Television Research Program at MIT, which I 
headed from 1983 until I retired in 1990, Russ Neuman, now at U Penn, 
directed an audience study at the Liberty Tree Mall in Danvers, Mass.  
Viewers were enticed out of the Mall with a small gift certificate and asked 
to look at TV and listen to audio and answer some questions.  The operation 
was completely anonymous; the viewers did not know it was MIT and were not 
told what they were observing.


Among other things, we compared "studio quality" 1125 I to NTSC with a variety 
of subject matter.  (There was only a slight preference for HDTV and little 
willingness to spend much for it, but this is not the point I am trying to 
make here.)  Even with very careful professional attention, it was very hard 
to get the picture contrast, brightness, and color close enough on the two 
displays so that an unbiased judgment could be made as to quality as 
dependent only on resolution and aspect ratio.  (The two pictures, when 
compared side-by-side, were the same height but different widths.)  I had 
the feeling that turning any knob that affected picture appearance on either 
display could have made either picture look better than the other.


Note that we controlled for viewing distance and for many personal 
characteristics of the viewers.  We also showed the pictures one by one and 
asked for a comparision with viewers' TVs at home.  One conclusion from this 
part of the study was that picture quality at home was controlled primarily 
by analog channel impairments, not scanning standards or bandwidth.  
Everybody thought that the pictures we showed were much better than they had 
at home.


Of all the quality factors studied, the most important was picture size.  Most 
viewers preferred larger pictures even when they said the smaller picture 
was sharper.


Temporal Aliasing


Another part of the ATRP study was motion rendition.  In a few days, I shall 
post citations to two PhD theses on this subject, which may be ordered from 
the MIT library.  I have no doubt that, at typical motion speed, temporal 
aliasing is always present at any reasonable frame rate.  Cinematographers 
know this and adjust camera technique appropriately.  Video people, who are 
really using 60 fps in NTSC, can move the camera much faster without causing 
much damage to the picture.


In work by Ed Krause, now at Imedia in SF, a demo was made testing various 
ways to render motion.  One was a computer-generated "bouncing ball" demo, 
in which 3/2 conversion to NTSC was compared with motion-compensated 
interpolation.  The latter won, hands down.  In another, video was made of a 
close-up of a wristwatch, causing it to move horizontally with sinusoidally 
varying velocity.  We compared frame repetition, motion-blur, and motion 
compensation. Again, the latter won easily.  Motion blur did not appreciably 
reduced the motion judder of frame repetition.  Note that when the eye is 
tracking motion on the TV screen, any attempt to reduce judder by blurring 
greatly reduces perceived sharpness.  Some video is deliberately shot with 
very short exposure time per field just to retain sharpness of moving 
objects.  ("You can see the stitching on the baseball!")


Krause also converted both 24-fps and 12-fps simulated film to NTSC using 
motion compensation.  The results were very good, altho the latter was not 
quite as good as the former, which was essentially perfect. Note that if 
motion compensated interpolation is to be used at the receiver without 
transmission of motion vectors, then the picture should not be blurred at 
all before transmission.


In another thesis by Dennis Martinez, motion-compensated interpolation was 
used to convert any frame rate to any other frame rate.  The subject was a 
closeup of a person's face while speaking.  (The audio was also successfully 
converted with duration changes of 20% or so.)  The results were essentially 
perfect with large changes in frame rate and corresponding changes in the 
speed of motion.