[tig] More Viper Stuff
Tue Jul 1 21:21:00 BST 2003
Steve Shaw wrote:
> Of interest the Dalsa camera captures 13 bit linear and can be set to
output true .CIN data so things are changing. What I wonder will the Sony
F950 be capable of outputting?
Just one point. We could easily start down a road of "more bits is better".
Reality is often much more complex than that.
For starters, the Dalsa camera uses a single focal-plane sensor with a color
filter array (CFA) generally referred to as a "Bayer Pattern" filter. This
is widely used in single chip video and still cameras. The Bayer CFA takes
a block of four pixels and pust two green filters, one red and one blue in
front of the respective pixels. The raw data off the sensor is useless
without running through a "de-mosaic" algorithm --a low-pass filter,
really-- to come-up with the missing color information. This is not unlike
going from 4:0:0 to 4:2:2 or 4:2:2 to 4:4:4, same sort of technology
involved, FIR filters. There are more elaborate algorithms out there that
attempt to produce better results, however, the real issue is that all of
these solutions (at least the one's I've seen) work great for still
photography ... the minute you introduce moving images things can get
strange. The problem is that the algorithms will produce different results
based on the frequency distribution (detail) on a given frame. As you play
through a sequence of frames the artifacts move!
Anyhow, so, the de-mosaic and filtering process (which must be done in post,
BTW) brings to the table questions/issues. One of them is: what is the true
resolution of an X x Y pixel sensor when, effectively it takes 2 x 2 pixels
to gather all the color information to synthesize one pixel. In addition to
that, the sampling of the colors in a CFA is spatially displaced: the red,
green and blue samples for a pixel are taken at different positions.
More specifically on A/D's. No information has come out on the
specifications of the imager for this camera, so it is hard to quantify any
of it. That's a problem with all contenders for the digital-cinema title,
BTW. They need to get off their broadcast mentality and provide real data.
The imager is now the emulsion. How would people feel if Kodak did not
publish technical data on their films?
Let me make an educated guess and place the pixels' full-well capacity at
80,000 electrons. Let' also place the noise floor at 20 electrons, which is
typical for a good CCD. This places the dynamic range at about 72dB, better
the 68dB published in their brochure.
A good indication of first-order A/D steps required is to simply divide the
full-well capacity by the noise floor. In this hypothetical example
80,000/20 = 4,000. A 12 bit A/D (4096 steps) is probably adequate,
particularly if you consider other noise sources.
To look at it a different way, if you digitized with a 12 bit A/D you are
representing 19.5 electrons per A/D step. Up that to a 14 bit A/D and each
A/D step is equivalent to 4.9 electrons. Unless you want to digitize noise
for safekeeping I'm not sure this is necessary.
Also, depending on operating temperature, you'll have a self-generated dark
signal (electrons that will start filling the pixel, even without light
exposure) that could be as high as 2 electrons per frame (50 electrons per
pixel per second).
A further issue --at least for me-- is that, in order to get all of that
data off the chip their current architecture uses twelve (yes 12) A/D
converters. I don't know about you, but having been around pretty high
priced equipment for 20+ years I think it is difficult enough to keep THREE
A/D's tracking perfectly, much less 12 ... particularly if you are not
operating said equipment in a temperature-controlled machine room.
Well, my point is that the number of bits per channel is only superficially
important. We've all seen stunning images on 8 bit RGB computer displays.
The question is where and how the number of bits in question where gotten
and just how true-to-life (as opposed to synthesized) they are.
Here's a link to a fairly high-end Kodak CCD for a point of reference. It
can't do video/cinema frame rates, but is otherwise state of the art.
Apogee Instruments has a nice little intro to CCD technology and the issues
involved: http://www.ccd.com/ccdu.html It is generally geared towards
scientific imaging (generally low frame rates) but is otherwise applicable.
Here's a nice tutorial, again, geared towards Astronomy, but otherwise
For a deeper look into imaging technologies I highly recommend visiting
eCinema Systems, Inc.
martin at ecinemasys.com
ecinema at ieee.org
More information about the Tig