[opendtv] Re: 3D compressed formats
- From: Mark Schubin <tvmark@xxxxxxxxxxxxx>
- To: opendtv@xxxxxxxxxxxxx
- Date: Fri, 17 Dec 2010 13:29:36 -0500
Hoo-Boy! What a can of worms!
There are literally dozens of 3D compression formats. You can download
a poster of some of the pros & cons, as described by Ateme, here (it's
the lower left): http://www.ateme.com/news_events.php5?Arg=153
Before I get into all the options, let me start by saying that almost
all of the 3D being transmitted today uses a straight side-by-side
squeeze: both views squeezed horizontally by a factor of two and placed
side-by-side in a single frame. That's one of the HDMI 1.4a formats.
The compression of those side-by-side images is standard.
Of course, there's a loss of 50% of the horizontal resolution, so Dolby
recommends a variation in which the left-eye image takes alternating
columns of pixels starting with the leftmost, and the right-eye-image
takes alternating columns of pixels starting with the next column. That
way, for portions of the image with no disparity, there's at least the
possibility that the two eyes' resolution can be additive (although,
once the temporal element of shuttered glasses is added, all bets are off.
Although that's the most common, it's just one form of
"frame-compatible" compression. In side-by-side, alone, in addition to
the simple squeeze, there are: squeeze & mirror; rotate & squeeze; and
rotate, squeeze, & mirror. Then there are side-by-side versions based
rotated 720p in a 1080p frame, thus losing both horizontal & vertical
resolution, but less of each, and there are mirrored versions of those,
too. There is also over-under (with mirrored version), alternating
field, alternating frame, column interleave, line interleave, and
quincunx, plus the Sisvel tile format (a full-sized, left-top, 720p
image in a 1080p frame with the right-eye view broken into pieces
surrounding it), something that allows set-top boxes to deliver 2D from
a 3D transmission without doing image processing. Finally (in
frame-compatible compression), there is anaglyph, with MANY color-pair
combinations, of which the most popular are red-cyan, green-magenta
(Trioscopics), and blue-amber (ColorCode 3D & SpaceSpex).
That brings us to non-frame-compatible (sometimes called service
compatible). One form takes any of the (non-anaglyph) versions above
and adds a helper signal for restoring the missing resolution. Then
there is 2D+delta (the difference signal representing the disparity of
the two views), 2D+delta+graphics, 2D+depth, 2D+depth+graphics, and
2D+depth+graphics+graphics occlusion. The +depth formats are useful for
multiview autostereoscopic displays.
Then there are various versions of MVC.
All of these are being experimented with, but the VAST majority of
current 3D is simply squeezed side-by-side feeding a single video
channel to an ordinary compression encoder. According to recent press
reports, the frame-compatible tile format is also picking up steam.
You can find more in slides 106-111 of one of my 3D tutorials, which may
be downloaded here:
http://schubincafe.com/blog/2010/11/more-than-you-ever-wanted-to-know-about-3d/
TTFN,
Mark
On 12/17/2010 9:46 AM, Mike Tsinberg wrote:
What are compressed video formats that are used for 3D broadcasting on
cable and satellite? Is there a formal definition or description of
compressed 3D format? HDMI has described 3D only in baseband.
Mike Tsinberg
Other related posts:
