[opendtv] Re: News: Intel introduces chips designed to improve Internet video quality
- From: Craig Birkmaier <craig@xxxxxxxxx>
- To: opendtv@xxxxxxxxxxxxx
- Date: Tue, 20 Nov 2007 14:08:20 -0500
At 12:42 PM -0500 11/20/07, Manfredi, Albert E wrote:
Craig Birkmaier wrote:
Note that there is no need for a 4:2:2 color space for progressive
images.
Can you explain this?
I thought that 4:2:2 uses half the luminance bandwidth for Cb and Cr,
horizontally. And that 4:2:0 uses half the luminance bandwidth for
either Cb or Cr, where Cb and Cr are either alternating line by line or
centered between alternating lines.
I thought that 4:2:0 was especially well suited for interlaced images,
and 4:2:2 would be ideal for progressively scanned images.
Mark has already addressed this to some extent, but there is a bit
more to the story.
Here are two sites with explanitions with accompanying graphics:
http://en.wikipedia.org/wiki/Chroma_subsampling
http://www.larryjordan.biz/articles/lj_sampling.html
The big problem here is that the notation system kinda breaks down
for progressive images.
The basis for all of this is that the human visual system is less
sensitive to color resolution than luminance resolution - you can
verify this by the distribution of foveal sensors that are
responsible for color vision.
An easy way to get an extra compression kick in composite video
system is to reduce the resolution of the color difference signals
prior to analog (NTSC/PAL) encoding. Thus the color bandwidth for PAL
is half that or the luminance, while in NTSC one color difference
signal gets half resolution while the other gets less than one
quarter.
Wikipedia tells us:
The YIQ system is intended to take advantage of human color-response
characteristics. The eye is more sensitive to changes in the
orange-blue (I) range than in the purple-green range (Q) - therefore
less bandwidth is required for Q than for I. Broadcast NTSC limits I
to 1.3 MHz and Q to 0.4 MHz.
In an ideal world, you could reduce the color bandwidth by half in
both the horizontal and vertical domains, as this is how the foveal
sensors are distributed in the human visual system.
But, as Mark indicates, reducing the color resolution in the vertical
domain is very difficult with interlaced imagery, as the alternating
lines of color information come from different sampling periods
(fields). To help maintain color integrity in digital sampling
systems for interlaced video we typically sample every line for both
color difference signals. The 4:2:2 notation tells us that we have
half the horizontal resolution per line for each color difference
signal.
The best way to think of this is that each field has 640 x 240 luma
resolution and 320 x 240 color resolution for each color difference
signal.
If we throw away vertical color information, as with MPEG-2 encoding
or the 4:1:1 sampling used in some low cost camcorders, we can
observe the loss of color detail. The Wikipedia article has some
excellent graphics to illustrate this.
the 4:2:0 notation has different connotations for interlaced and
progressive video frames. As i noted, the notation system loses its
meaning when we try to apply it to progressive frames. When we
subsample the color difference signals in the vertical domain for a
progressive image we typically wind up with 1/4 the total number of
color difference samples, but the color image has 1/2 the resolution
in both the H & V axis. This is possible because all samples come
from the same temporal sampling period.
When we try to apply MPEG-2 encoding techniques to interlaced video
we run into problems. The luminance samples for the combined frame
come from distinct fields that are 1/60th of a second apart in terms
of the sampling time. When there is a high level of motion this can
cause significant problems for the algorithm; thus there are routines
in MPEG-2 to encode interlace that allow the fields to be compressed
independently when there is rapid motion. But the color difference
samples that correspond to the luminance samples come from both
fields and must be combined to reduce the vertical resolution by
half. This results in a significant loss of color resolution for
interlaced sources with MPEG-2, as can be seen in the images on
Wikipedia.
In the same vein, using 4:2:2 sampling for progressive images only
brings up the level of horizontal detail while doubling the sampling
rate for the color difference signals. The nominal way to deal with
color subsampling with progressive frames is to retain full
resolution (4:4:4)
or to reduce color resolution by half with the odd nomenclature of
4:2:0. We could also reduce it by half again, with a more logical
nomenclature of 4:1:1 (which is also different than sampling
interlace with at 4:1:1. I might add that we are also starting to see
the 8:4:4 nomenclature, which is typically used to denote a higher
sampling rate for the luminance signals.
So in essence your thinking was just reversed:
4:2:0 is best for sampling progressive images
4:2:2 is best for sampling interlaced images
Another reason why there is no logical rationale for using interlace
when we move to digital video compression.
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:
