|
[opendtv]
||
[Date Prev]
[06-2007 Date Index]
[Date Next]
||
[Thread Prev]
[06-2007 Thread Index]
[Thread Next]
[opendtv] Re: Red camera lenses
- From: Tom Barry <trbarry@xxxxxxxxxxx>
- To: opendtv@xxxxxxxxxxxxx
- Date: Sun, 03 Jun 2007 21:57:41 -0400
Mark Schubin wrote:
>> I'm guessing you might divide the H & V resolutions by 1.414 since it
>> is effectively a 45 degree rotated grid.
>
> No (unless you're talking about Sony's Clear-Vid sensors or the ones in
> Fuji digital still cameras).
>
What I meant on the rotation applied only to the green pixels. A typical
Bayer pattern looks like this in, say, an 8x8 square (hope it displays
right)
g r g r g r g r
b g b g b g b g
g r g r g r g r
b g b g b g b g
g r g r g r g r
b g b g b g b g
g r g r g r g r
b g b g b g b g
so it seems pretty straightforward the b and r components have only half
resolution in both the H an V directions. But out of 64 pixes 32 of
them are green. I like to think of sampling densities per area, samples
per unit^2. This gives an effect square root of 32 or 5.66 effective
resolution both horizontally and vertically for the green pixels.
To see how this density might be achieve, imagine (for green sampling
only) rotating the entire grid 45 degrees, like:
g
b r
g g g
b r b r
g g g g g
b r b r b r
g g g g g g g
b r b r b r b r
g g g g g g g
b r b r b r
g g g g g
b r b r
g g g
b r
g
Pretend the above was the same square as the previous one but rotated up
on edge. Surely rotating a camera doesn't change the resolution. Now
the g samples are in close packed rows and the sampling density is
greater for Nyquist purposes. Of course we now have the complication of
oversampling and filtering along both diagonals instead of rows and
columns but that can be done. And it increases the effective green
resolution as I state before by about 1.414.
Except I don't think you can really use that resolution anyway, Nyquist
be damned, because it seems without filtering in an RGB setup the three
different colors would have to have the same resolution or you would
maybe have high frequencies chroma shifts every other pixel. The RED
camera claims to have an optical low pass filter to avoid moire problems
and I'll bet there really is no extra green resolution over r & b.
Wonder if they could get around this by making g mean neutral grey,
making making more of a YUV pattern.
>> But I'm only starting to look into this.
>
> Enjoy! Many of the great minds in the industry also haven't agreed on
> the results. ARRI, for example, posits a significant resolution loss
> due to the color filtering; Dalsa doesn't.
>
I'm actually puzzling why Bayer is better. Its advantage seems to be
the single pixel color filters allowing a single grid to sample
everything without prims, etc.
But, except for the manufacturing problems of non-square grids it seems
to me a triangular pattern would be better without the extra green
resolution problems I mentioned above. Does anybody sample in close
packed triangles with single pixel color filters, like:
r g b r g b r g
b r g b r g b
r g b r g b r g
b r g b r g b
r g b r g b r g
b r g b r g b
r g b r g b r g
b r g b r g b
I would assume the output would be soon converted to rows and columns
but it seems it might be useful to first sample that way.
- Tom (not a great mind, but a curious one)
Tom Barry wrote:
I have only been skimming the recent discussion of lenses since I know
very little about them. But I'm somewhat enamored by the rumors of
the new RED 4K camera and today was on their forum reading the RED FAQ.
I happened to notice it stated:
"Since Mysterium™ captures all three color wavelengths at the same
depth [unlike a 3-CCD system], standard lenses designed for film
cameras [both cinema and still photography] can be used. More on this
in the lens section below."
Dunno if this factors much into the lens discussion or not, but it
seemed it might, and the darned camera sounds quite nifty.
Here are the two ways it factors into the lens discussion. Most
professional video cameras use a prism for color separation. That adds
a back focal length of around 100 mm (it varies with imager-size
format). Movie (and still) cameras tend to have much shorter back focal
lengths. That prevents those lenses from being used directly on video
cameras.
Without changing the color-separation system, one can use some form of
relay optics. Some (e.g., P+S Technik, Red Rock Micro) create a real
image from the film lens on a ground-glass screen and shoot the screen
for the video camera. They tend to rotate the ground glass so it
doesn't create a fixed pattern in the image. Others (e.g., JVC,
Century) relay a virtual image from the film lens to the video camera.
It is also possible to change the color-separation system. There have
been three techniques for this: rotating color filters (used in the old
CBS-color system, the Colorgraphics system, and the Apollo mission
camera), stacked sensors allowing higher-energy colors to pass through
the layers intended for lower-energy colors (the Foveon sensor used in
Sigma cameras), and on-chip color filtering -- stripes (Abtography, the
Panavision Genesis, old inexpensive color video cameras) or on a pixel
basis (Bayer, diagonal, or the lower-ratio systems used in some of
Sony's Clear-Vid systems). The color filters are most popular: stripes
in the Panavision Genesis and Bayer in the ARRI D20, the Dalsa Evolution
and Origin, the EasyLook Modula HD, the NAC Memrecam fx K4, the Red One,
the Silicon Optics SI-2K, SI-2K Mini, and SI-1920, the Vision Research
Phantom HD and Phantom 65, the Weinberger Cine Speedcam, the Weisscam
HS-1, and many smaller-format cameras as well as cameras not intended
for TV use.
The other issue relates to longitudinal chromatic aberration. In a
prism-based system, it is possible to mount the three color sensors at
slightly different back-focal distances; that is not possible in
single-sensor systems, though it is at least theoretically possible to
change the refractive characteristics of the color filters to
compensate. The adjustment for longitudinal chromatic aberration was
much more common in the old days of tube cameras with wide variations
between lenses.
The whole RED FAQ on their forum can be seen at:
<http://www.reduser.net/forum/showthread.php?t=1487>
I think I'll have to learn more about Bayer patterns.
They're fairly simple. In each square of four individual-pixel sensors,
the upper left is green, the upper right is blue, the lower left is red,
and the lower-right is green. The order can vary, but all adjacent
four-sensor squares must be the same. Thus, there will be a
green-blue-green-blue row and a red-green-red-green row. When the
adjacent squares swap blue and red, the result is a diagonal filter
instead.
Anybody know what the effective Nyquist sampling limit is on a Bayer
pattern.
It depends what you're measuring. If you're looking at alternating
white-black patterns, the color filter may be largely ignored (luma
variations are compensated in post-imager processing). If you're
looking at color patterns, it would also depend on the pattern. If you
assume an unfiltered optical system that exceeds the resolution of a
Bayer-filtered imager, you could have, say, a blue horizontal or
vertical line that would disappear in one position and reappear one
pixel-shift later.
I'm guessing you might divide the H & V resolutions by 1.414 since it
is effectively a 45 degree rotated grid.
No (unless you're talking about Sony's Clear-Vid sensors or the ones in
Fuji digital still cameras).
But I'm only starting to look into this.
Enjoy! Many of the great minds in the industry also haven't agreed on
the results. ARRI, for example, posits a significant resolution loss
due to the color filtering; Dalsa doesn't.
TTFN,
Mark
----------------------------------------------------------------------
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.
--
Tom Barry trbarry@xxxxxxxxxxx
----------------------------------------------------------------------
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.
|
|
