[opendtv] Re: State of the Art LCD 45" HDTV

At 7:56 PM -0400 10/27/04, Tom Barry wrote:
>You must have very good eyes.  I've usually figured that over 4 screen
>heights viewing distance 720p was fine, reserving 1080i/p for less than
>that.

Actually I was just being lazy. I did not want to take the time to 
calculate the cycles per degree of potential resolution available at 
the viewing distances I mentioned. And there are other factors that 
complicate this discussion.

As I noted, the samples per inch of the display works out to be 49 
samples per inch. Personal computer displays, which are nominally 
viewed at a distance of about 30 inches typically have a DPI (same as 
samples per inch) between 72 and 96. This suggests that the optimal 
viewing distance is somewhere in-between that needed for a computer 
display and that needed for a TV display. I guessed that it would be 
4-5 feet (48-60 inches), which is clearly much closer than most 
people would sit to watch TV ( but conceivable for viewing of an 
information display in a signage application).

At a more typical TV viewing distance of 7-10 feet this display is 
overkill - it would not be possible to resolve the available detail. 
Tom is correct, that 720P is more than adequate for a display of this 
size when viewed at the "Lechner distance."\

>But I've been wondering lately if I have to amend that a bit for fixed
>pixel displays with well defined pixels and no implicit filtering from
>CRT spot beam size.  The fixed pixel displays may well create some
>artifacts that were not an issue back when the original viewing and
>visibility tests were mostly performed.

An interesting observation, but there is no need to amend your thinking.

The main difference with fixed pixel displays is that they are 
capable of presenting detail that is not acceptable for video or 
natural images, which require Nyquist filtering of the samples to 
prevent the perception of aliasing. We can illuminate alternating 
lines at full intensity - i.e. black white transitions. This ability 
to present non-Nyquist filtered samples is critical when you want to 
enhance contrast at the expense of some aliasing, as is the case with 
the text you are reading now.

But every fixed pixel display can also present properly Nyqist 
filtered samples, at the expense of perceived resolution and 
contrast. You actually see more accurate image portrayal than with a 
CRT because the samples are always in the right physical location and 
the sample size (as compared to the spot beam size) is uniform 
everywhere on the display (note that this may not be absolutely true 
for projection systems because of the potential for lens distortions, 
key stoning, etc.

The other obvious difference is that the "trick" that makes interlace 
work on scanning displays, does not work with "constant illumination 
displays." The human visual system integrates the scanning spot to 
create the perception of moving images. By interlacing we sacrifice 
vertical resolution in order to improve temporal resolution - the 
human visual system integrates the constantly changing half vertical 
resolution fields. This works quite well  if we use enough filtering 
to hide the artifacts of interlacing.

But with displays that illuminate entire video frames continuously, 
for most or all of the temporal sampling period, the "trick" doesn't 
work. We must create the information missing from the interlaced 
sampling process to create complete frames. This can be done crudely 
with spatial/temporal filters that hide the artifacts, or we can use 
complex motion compensated prediction techniques to "predict" the 
missing samples. The latter will typically produce more detailed 
images, but suffers from problems with many types of imagery that are 
very difficult to predict - for example;
- When new information is revealed between temporal samples (e.g. a 
rotating globe).
- Plastic deformations - i.e. reflections off of non-planar surfaces
- New information entering the edges of the frame

Some of these issues can be improved with more temporal samples in 
the prediction buffers. Some require complex calculations akin to 
those used for the rendering of complex 3D animations.

Life is SOOOOOOO much easier when we sample images properly in the first place.

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: