[opendtv] Re: Math of oversampling - a REALITY CHECK
- From: Craig Birkmaier <craig@xxxxxxxxx>
- To: OpenDTV Mail List <opendtv@xxxxxxxxxxxxx>
- Date: Mon, 2 May 2005 08:18:53 -0400
I mistakenly sent this message only to jeroen yesterday. He has since
covered some of wthe issue I write about here.
Craig
At 2:24 PM +0200 4/29/05, Jeroen Stessen wrote:
>Hi,
>
>See below ! (Was not posted.)
>
>> To be honest, it took me a few moments to see the
>> difference between the 576P and 720P images.
>
>You will now !!
>
>Be careful that you don't see only what you want to see.
>It depends on the image content, and on the display, but
>maybe also on any prejudice that you might have...
>
>BTW, this 20" CRT monitor at 1600x1200 is rasor sharp, better
>than any LCD that I might find in this place. Don't worry.
>- Tom
Let's take the prejudice out of it and consider the realities of
delivering high quality images to millions of consumers who will
watch their expensive new TVs in a manner that is not the same as
that we are using to evaluate these images.
To this point we have been looking at only a single important factor.
Jeroen's latest images do a good job of demonstrating the information
carrying capacity of rasters at various spatial resolutions. As THEY
SHOULD, we see more detail as the density of the raster is increased.
But it is important to remember that we are NOT using real world
analysis techniques.
We are looking at a single frame of information for extended periods
of time on a high resolution display viewed at a close distance,
where we can change the ability to see small details simply by moving
our heads in and out over a range of a 15" to 30" viewing distance.
Some of us are even zooming into the image to see the underlying
sampling structure.
Here is a simple test that tells the real story. Open up the 1080P
file and place it in a window on the top half of your high resolution
display - make certain that it is displaying the samples in a
one-to-one relationship with the display samples (100% or actual size
depending on the viewer you are using). Now open one of the lower
resolution files, I used the 576P file, and place it below the 1080P
file so that the same region of the image is visible in both windows.
What do you see. The answer depends on the distance your eyes ar from
the display.
At a viewing distance of 15 inches the 1080P image looks
significantly sharper. At a distance of 30 inches it is still
sharper, but the difference is much smaller. At three feet, most of
us will not be able to tell the difference.
What we are talking about here is display scalability. There are
three critical factors at work here:
1. The actual resolution of the display - with modern "lithographed'
display technologies the actual number of pixels is important, but
there may be other factors that influence the delivered resolution,
including:
a. any effects from optical filters (the lenses and screens used
for projection systems;
b. Internal image processing circuitry - the bandpass of the
amplifiers, and any
inter-format conversions that may be taking place;
c. Diaplay gamma and related issues of how the display generates
and maps pixel
values to the screen;
c. User adjustments to contrast, brightness and sharpness that affect the way
samples are mapped to the display.
2. Screen size - How big are the individual pixels? For a panel
display that has a 1280 x 720 raster, the pixels are much larger on a
40" screen than a 20 inch screen. Generally speaking larger screens
need more samples, but this is directly related to the third
criteria, the viewing distance. Those big stadium displays often have
pixels that are measured in inches rather than millimeters.
3. Viewing distance - how far is the viewer from the display. I noted
above that it is easy to cut the viewing distance to a computer
display by half, simply by moving one's head in and out. At the
Lechner distance (~9 feet) moving one's head has little if any impact
on the ability to resolve more detail. The process of choosing a
display with the appropriate level of resolution BEGINS by
determining the designed viewing distance for the application.
What all this boils down to is that one can accurately determine the
resolution requirements for ANY display application if we know the
screen size and viewing distance. What we are seeking is to have the
display match, or slightly exceed the visual acuity of the viewer in
order to deliver a sharp image without the perception of artifacts,
including the ability to see individual lines and/or samples in the
displayed image.
We also need to factor in the reality that for entertainment viewing,
the designed viewing distance, measured in multiples of screen
height, decreases as the display subtends a larger portion of our
field of view. On average we tend to sit 7-10 picture heights from
small "low resolutions displays such as NTSC and PAL televisions. For
720P displays between 40 and 60 inches diagonal we tend to sit 4-6
picture heights. For 1080P displays we may sit 3-4 picture heights
when the screen size exceed 100 inch diagonal, but will typically sit
further away from smaller screens.
This is the crux of the question originally posed by Bob Miller. How
much resolution do we need to deliver through the emission system to
create the perception of a sharp picture on the average range of
screen sizes that will be used to view the content?
The reality is that 576P is adequate for most of the screens that are
being purchased in Europe and the U.S. today. You need a screen
larger than 40" diagonal before 720P delivers a perceptible
advantage. And you will see NO DIFFERENCE between 720P and 1080P
until the screen size is larger than 70 inch diagonal at typical
viewing distances.
But there is another critical factor that must be considered as well.
What is the quality of the samples that are actually delivered
through the emission system? have they been impaired by the
compression system, as we saw in the images that Jeroen sent to us?
At NAB there were three companies showing very sophisticated image
processing/de-interlacing chips that are capable of outputting
1080@60P images to high resolution displays. These processors do a
good job with interlaced SDTV source. They do a fantastic job with
progressively scanned 720P source. This leaves us to consider where
we need to optimize the system.
For the vast majority of consumers, 1080P is overkill. Even 720P is
beyond the requirements for the displays that will dominate the mass
markets. Given the same channel bandwidth (say 10-12 Mbps) and H.264
compression, which will deliver the highest quality samples to the
image processor inside all of those new consumer displays:
1. 1920 x 1080 @ 60i
2. 1280 x 720 @ 60P
The answer is obvious. We need to focus on using digital compression
tools properly to deliver high quality samples, and let powerful
image processing chips deal with the up conversions for really big
screens.
Regards
Craig
P.S. It is also important to note that the "Philips" image" that
Jeroen used in this test was not a frame from a sampled video image.
It was composed in a computer - probably at a resolution
SIGNIFICANTLY higher than 1920 x 1080, then output to film for
printing purposes.
----------------------------------------------------------------------
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:
