[opendtv] Re: "we'll forever be stuck with by going ATSC"

  • From: Craig Birkmaier <craig@xxxxxxxxx>
  • To: opendtv@xxxxxxxxxxxxx
  • Date: Fri, 14 Jan 2005 14:00:27 -0500

At 8:42 PM -0500 1/13/05, Tom Barry wrote:
>Darn it Craig, that is just silly.
>
>You keep going on about 1970's studies that claimed that, compared to
>the expectations of the day, 22 cycles per degree of resolution seemed
>"very sharp".  And yet your same article said people might be able to
>see up to 40.
>
>"Very sharp" is not the best there is.  The best occurs when there are
>no visible artifacts (including obvious image softness) even when you
>sit close enough the screen subtends at least 30 degrees of your visual
>field.

Sorry Tom. It may sound silly, but it is factual. Perhaps I need to 
do a better job of explaining...

First, some of these studies are MUCH older than the '70s. Scientists 
have been studying the human visual system for centuries and the 
notion of measuring visual acuity in terms of cycles per degree 
probably goes back to the 19th century. What is important here is to 
understand the relationship between screen resolution and human 
perception.

When NTSC was designed, it was well understood what it takes to 
present an image to the human observer that is perceived as being 
sharp. Human perception is a very complex subject with multiple 
parameters that must be considered. Staring at a high resolution 
still image versus a video stream is a good case in point.

With a still image you can focus on a particular area of the image - 
assuming it is not vibrating or moving, your foveal vision will be 
able to stay focused on fine details long enough for a high 
resolution view to be constructed in your brain. This takes several 
hundred milliseconds.

A good way to look at this from an information theory perspective is 
that you need only collect that highly detailed view once. Extending 
this to video compression, since there is no temporal element, you 
can send the frame once and you are done.

With a video stream we add a temporal dimension, and this changes 
everything. You can focus on fine detail in one area of the image, 
but the samples are likely to keep changing, literally in front of 
your eyes. If the display size and viewing distance are well matched 
to the central foveal image receptors, as was the case for NTSC, the 
human visual system can capture the entire image without having to 
dart around to capture high resolution views. It is not a coincidence 
that the <11 degree field of view for NTSC is very close to the area 
that can be captured by foveal receptors. When we extend that field 
of view we now start stimulating extra foveal receptors which direct 
eye movement. A big part of the induction effect for HDTV is related 
to covering that wide field of view  so that you are forced to use 
directed eye movements. Unfortunatley induction typically does not 
occur if the viewer is very close to the display - this may have 
something to do with the fact the the eye is forced to focus on the 
near field.

NHK did most of its HDTV research in the mid and late '70s. In most 
of the papers I have read, they used film based images to determine 
the correct parameters to stimulate the perception of a high 
resolution image AND the induction effect. One of the major reasons 
for this was to eliminate the effect of a scanning raster on the 
studies. After determining the resolution requirements they then 
calculated the number of lines needed to provide the desired level of 
detail, WITHOUT the perception of the scanning raster. It was 
determined that a progressively scanned image with 700 lines was 
sufficient. Unfortunately, they were designing this system for 
satellite distribution using MUSE, and there was NOT enough bandwidth 
to deliver something akin to 720P. So they resorted to the same trick 
used to deal with the bandwidth limits for NTSC - interlace.

You seem disturbed by the notion that some people can resolve fine 
details at resolutions as high as 40 cycles per degree. But there are 
some important caveats here. For most forms of natural imagery, the 
average viewer can resolve sin the range of 22 to 30 CPD. Above this 
average you need VERY high contrast for the eye to respond. My 
favorite example is a blinking red traffic light off in the distance. 
If it is daylight, you may not be able to resolve the light until you 
are fairly close - say 1/4 mile; if the sun is right behind the light 
you will not see the red light at all as it will be overwhelmed by 
the intensity of the sun. But at twilight, or after dark, you can 
perceive the light at much greater distances(i.e. your eye is 
reacting to a stimulus that may be in the range of 30-40 CPD. This is 
possible because of the high contrast, and the fact that the light 
itself is the source of that contrast. If the light is blinking, you 
may even perceive it sooner as the temporal element helps with 
perception.

Mark Schubin likes to talk about another aspect of this issue....the 
broken pixel.

What happens when there is one pixel that is permanently set to some 
value, say red? If the pixels that surround it are highly contrasting 
you may see that red pixel; if the contrast level is small, or the 
hue is close you probably will not see it.

One advantage of having samples that are smaller than you can resolve 
is that a bad sample will not be readily apparent.

But for normal conditions, an HD display need only provide enough 
resolution to satisfy the average viewer in terms of acuity. That is 
why i typically say 22 CPD or greater.

Now what happens when you increase the viewing distance relative to 
the designed viewing distance? For example, Bert and I sit down to 
watch his new 26" panel. Bert, to prove a point chooses to sit at 
three picture heights. I choose to sit at 5 picture heights. If our 
visual acuity is the same, he is going to be able to resolve more 
detail. At five picture heights some of the fine details will be too 
small for me to resolve.

This is where the decisions of the content producer come into play. 
If it is KNOWN that the viewers will all have high resolution 
displays that cover 30 degrees of the field of view, you can sit on 
wide shots and let the viewer decide what to focus on. But this is 
not going to work very well  for me, sitting at five picture heights. 
I will see a very sharp picture with a bunch of little ants running 
around on the football field.

If you watch a football game in NTSC, what do you see? Wide shots are 
used only to "establish." After that, everything is close-ups because 
of the limited size and resolution of the NTSC display.

This is just one of many reasons why it is difficult, if not 
impossible to shoot once and distribute to displays with significant 
differences in size and resolution. This is why HDTV is developing as 
a complementary service to SDTV.  IF the content does not need the 
extra resolution, as is the case for most motion pictures, sitcoms 
and episodic shows, you can get away with delivering the same content 
to displays with different capabilities, but when you are optimizing 
for and selling resolution you by necessity sub-optimizing for less 
capable displays.

>
>You also quote studies where people would try to move back to their
>habitual (in screen heights or furniture feet) viewing distance up until
>the point where they were presented with theater sized screens.  Please
>try the same experiment again with people already accustomed to viewing
>HDTV displays and tell me how it works out.


The studies of preferred viewing distance, these are much more 
recent. The studies I have been referring to took place in the late 
'80s and early '90s.  I suspect that if they were repeated today with 
the displays that are now available the results might be somewhat 
different. The major reason for this is that most of the studies were 
done with CRT based displays, both direct view and projection.  As we 
all know now, these displays typically do not offer the desired 
combination of brightness and contrast that we are now seeing with 
LCD, DLP etc. And they also introduce raster artifacts that impair 
the ability to resolve fine details. I agree that people may want to 
sit a bit close to a good HDTV display, but I would not expect a huge 
change in the results.

Similar "studies" are done everyday across America. Walk into any 
theater, and see how the seats fill up. The front rows are always the 
last to fill up. People are uncomfortable sitting too close to the 
screen.

So in the end, the correct way to buy an HDTV display is to start 
with the viewing conditions. That is, what will be the average 
viewing distance. From there is is easy to determine how large a 
screen you will need, and at what resolution to deliver what is 
perceived as a sharp image. And anything that is at 22CPD or greater 
will look sharp to virtually everyone.

I'm not certain that experience watching an HD display is going to 
change things either.  I have seen very few HDTV installations where 
the primary seating is at 3 picture heights. This is not surprising 
for several reasons;

Room size, furniture placement and screen size.

The majority of people are buying BIG screens because that is what 
HDTV is all about. Now we are beginning to learn how people will use 
small high resolution screens. I agree that the jury is still out on 
this one.


>You already have posted that the viewing distance you would use
>personally for picture quality is closer than your wife would allow
>based upon furniture dynamics & traffic patterns.  It seems likely you
>will adjust but, if not, then next time you will probably buy a larger
>TV.  But still not 100 inches.

A reasonable assessment. By the way, the furniture managed to creep 
back in while she was sleeping, and we are now talking about 
replacing the coffee table, which is too big and the cause of some of 
the traffic flow problems.

A larger screen is out of the question because of fireplace. The 50" 
screen is more than adequate for all of the possible viewing 
locations in the room. I am very comfotable with the 12 foot primary 
viewing distance for this screen (about 5 picture heights). If i get 
a Mac Mini that can drive the full 1280 x 720, it will be intersting 
to see how well the Mac GUI works at this distance.

>Very few have 100 inch TV's and yet many of use DO occasionally see the
>    WOW factor and reap the benefits of HD.  On sets no larger than yours.

Go back and look at what I have been writing...again.

A 100" screen is overkill for most homes. I have consistently 
indicated that the HDTV viewing experience begins with screens larger 
than 40".  A screen between 40" and 100" with 1280 x 720 resolution 
will deliver a sharp image to the viewer at viewing distances equal 
to or greater than 3 picture heights. This pretty well covers >99% of 
all installations. Only when the screen size exceeds 100" do you need 
additional samples to prevent the perception of the raster or a a 
soft image. This is why I believe 1920 x 1080 is a waste of time for 
ANY commercial DTV deployment.

Regards
Craig
 
 
----------------------------------------------------------------------
You can UNSUBSCRIBE from the OpenDTV list in two ways:

- Using the UNSUBSCRIBE command in your user configuration settings at 
FreeLists.org 

- By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word 
unsubscribe in the subject line.

Other related posts: