New Kodak Sensors See Well in Dark - claims a 1 - 2 f/stop advantage

New Kodak Sensors See Well in Dark - claims a 1 - 2 f/stop advantage.

See:

http://www.pcworld.com/article/id,13...s/article.html

David

David J Taylor wrote:
New Kodak Sensors See Well in Dark - claims a 1 - 2 f/stop advantage.

See:

http://www.pcworld.com/article/id,13...s/article.html

David


If it works as described, it just might fix my single complaint about my
wife's camera, poor low light operation. We shall see.

On Jun 14, 6:11 pm, "David J Taylor" -this-
bit.nor-this-part.co.uk wrote:
New Kodak Sensors See Well in Dark - claims a 1 - 2 f/stop advantage.

See:

http://www.pcworld.com/article/id,13...s/article.html

David

One could cynically say that it would just about bring them into line
with other manufacturer's sensors... )O;

But the article is interesting, and you have to ask why it hasn't been
done before - I've certainly wondered about the logic of having two
green pixels for every one blue/red.. Yes, I know about the eye's
sensitivity to green light, but even so....

I hope Kodak are still in the game by the time they get it to work (O:

wrote:
On Jun 14, 6:11 pm, "David J Taylor" -this-
bit.nor-this-part.co.uk wrote:
New Kodak Sensors See Well in Dark - claims a 1 - 2 f/stop advantage.

See:

http://www.pcworld.com/article/id,13...s/article.html

David

One could cynically say that it would just about bring them into line
with other manufacturer's sensors... )O;

Actually, Kodak makes some of the best CCDs around. They are
commonly used in scientific applications needing the highest
quality.

But the article is interesting, and you have to ask why it hasn't been
done before - I've certainly wondered about the logic of having two
green pixels for every one blue/red.. Yes, I know about the eye's
sensitivity to green light, but even so....

It is an interesting concept, adding a 4th pixel type
with no filter. That would increase the number of photons
collected by the unfiltered pixel by at least 3x.
The disadvantages include:

Lower color spatial resolution.

The unfiltered pixels saturate 3x or more before
the color pixels, so exposure times are shorter,
leaving the color pixels underexposed. Thus
increased chrominance noise.

Roger

On Thu, 14 Jun 2007 04:17:53 -0500, Ron Hunter wrote:

http://www.pcworld.com/article/id,13...s/article.html

If it works as described, it just might fix my single complaint about my
wife's camera, poor low light operation. We shall see.

Does "expected to be available in products in 2008" mean available
for sale in 2008, or available for testing by camera manufacturers
in 2008? It may be the difference between being able to buy
something using that sensor in 2008 and 2010. And if it's initially
made available in cameras such as Fuji's F10/F20/F30, even if it
fixes your single complaint about your wife's camera, if like the
F10/etc. it has no viewfinder, will you patiently wait for another
couple of years? g

If you think about Kodak's "new" sensor technology, it veers from
the Bayer design in the opposite direction that Foveon took. Maybe
we should call it a Super Bayer sensor, even if it wasn't theorized
or designed by Kodak's Dr. Bryce Bayer.

On Thu, 14 Jun 2007 05:03:35 -0700, Roger N. Clark (change username
to rnclark) wrote:

But the article is interesting, and you have to ask why it hasn't been
done before - I've certainly wondered about the logic of having two
green pixels for every one blue/red.. Yes, I know about the eye's
sensitivity to green light, but even so....

It is an interesting concept, adding a 4th pixel type
with no filter. That would increase the number of photons
collected by the unfiltered pixel by at least 3x.
The disadvantages include:

Lower color spatial resolution.

Isn't that also a theoretical drawback of Bayer vs. Foveon
sensors? If so, losing a bit more probably won't be too significant
compared to the advantages.


The unfiltered pixels saturate 3x or more before
the color pixels, so exposure times are shorter,
leaving the color pixels underexposed. Thus
increased chrominance noise.

That would be true if all of the pixels were the same size, and
even the use of different microlenses may have an effect. In some
of today's sensors not all pixels are the same size. I'm assuming
that Kodak's "panchromatic" (clear) sensor will much larger than the
others, so it won't saturate as quickly as the smaller pixels. Fuji
has done with their SuperCCD SR sensors which utilize two different
size pixels, the larger ones being the more sensitive. These are
used in Fuji's S3 Pro and S5 Pro, not to be confused with the older
SuperCCD III sensor used in the S2 Pro.

http://www.dpreview.com/news/0301/03...superccdsr.asp

Note that the above web page states that the two different sized
pixels share the same microlens. This was evidently changed for the
S3 Pro, where DPReview wrote :

The S3 Pro utilizes Fujifilm's "extended dynamic range" SuperCCD SR
sensor which features two photodiodes at each photosite (a single 'input
pixel'). The 'S' pixel has normal sensitivity and captures the same range
of light as a conventional CCD photosite, the 'R' pixel is smaller and has
a lower sensitivity and is designed to capture detail above the saturation
point of the 'S' pixel, the camera can then combine the information from
the 'S' and 'R' pixels to produce an extended dynamic range and avoid
the loss of detail due to over-exposure.

Initially the design of the S3 Pro's SuperCCD SR sensor was the same as
that used in the F700, F710 and S20 Pro (the R pixel above the S pixel;
diagram on the left below). However Fujifilm has since then slightly
modified the design by moving the R pixel to the empty space between
S pixels which now allows for a larger S pixel (diagram on the right below).
This also means that the S3 Pro must have a fairly unusual microlens
layout (as each S and R pixel must have its own microlens) and we'll be
very interested to see if this has any impact on image quality.

http://www.dpreview.com/reviews/fujifilms3pro/

In article , "Roger N. Clark (change
username to rnclark)" writes

It is an interesting concept, adding a 4th pixel type
with no filter. That would increase the number of photons
collected by the unfiltered pixel by at least 3x.
The disadvantages include:

Lower color spatial resolution.

That is certainly true with the colour matrix as shown on the article,
but I wonder if that is just a PC World graphic artist impression of
what the Kodak article describes or if it really is a direct copy of a
Kodak graphic.

One of the features of the original Bayer pattern is that every green
pixel has two red and two blue pixels as the nearest neighbours, while
every red and blue pixel has 4 green nearest neighbours and 4 of the
remaining colour as next nearest (ie diagonal). It is this close packed
nature of the coloured pixels that gives the pattern its high intrinsic
colour resolution. All of the colour information is contained in a 2x2
pixel cell.

However, that close packing is lost in the "high sensitivity pattern"
shown in the article, which is a repetition of this 4x4 unit cell:

WBWG
BWGW
WGWR
GWRW

For example, each red pixel has 4 white nearest neighbours with two
green, one blue and a RED diagonal. Even worse, each green pixel has 4
white near neighbours and one each of red and blue and TWO green on the
diagonals.

You need a complete 4x4 array of pixels to guarantee you get all of the
colour information, compared to a 2x2 pixel array for the Bayer matrix.
As you point out, this produces extremely poor colour resolution, as
seen by the double red and blue pixels on the diagonals.

However, it doesn't have to be that way, which is why I suspect a PC
World graphic artist screw up.

I think that an improved pattern (possibly the correct pattern) should
be a repetition of this 4x4 unit cell:
GRWR
BWBG
WRGR
BGBW

If this isn't what Kodak meant, then I claim this pattern as the
"Kennedy Array" ;-)

There are several differences between this matrix and that shown in the
article:
Every green pixel has two red and two blue nearest neighbours (as Bayer)
Every white pixel has two red and two blue nearest neighbours
Every red or blue has two green and two white nearest neighbours and
four of the remaining colour as the next nearest (diagonal).
No red or blue pixels are adjacent to pixels of the same colour (as
Bayer)

The result of this is, of course, that it has almost the same colour
resolution as the original Bayer matrix, with all of the colour
information being contained in any 2x2 pixel array.

Of course it doesn't have the same increase in sensitivity of the matrix
shown in the article - in each 4x4 block the original article there are
8 white pixels, whilst in my revision this is reduced to 4, but I think
that would be a preferable compromise to what PC World (not exactly the
world leaders in imaging technology) show. ;-)

The unfiltered pixels saturate 3x or more before
the color pixels, so exposure times are shorter,
leaving the color pixels underexposed. Thus
increased chrominance noise.

Agreed, that's what would happen if used like that all the time, but I
don't think that is what is intended. Exposing for unsaturated white
pixels would only occur in low light levels, when the shorter exposure
times are necessary (that's what Kodak say they are aiming this at in
their text). As you say, that would result in an increase in chroma
noise - but improved luminance noise over a conventional Bayer array. We
know that the eye is less sensitive to chroma noise - that was the
essence of colour TV until recently and is also a feature of the Bayer
sensor itself. So this could be a significant improvement overall at
where light is limited. However, at the "normal" apertures and shutter
speeds used for the Bayer array, the unfiltered pixels could be allowed
to saturate (which would only happen in certain parts of the image
anyway) and then simply filtered out in the processing with very little
impact on the image resolution or colour.

In fact, it would have even less impact on the image with the "Kennedy
Array" where not only is all of the colour information contained in any
2x2 pixel array, but all of the low light white signals are as well. ;-)
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's ****ed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)

On Thu, 14 Jun 2007 04:17:53 -0500
Ron Hunter wrote:

David J Taylor wrote:
New Kodak Sensors See Well in Dark - claims a 1 - 2 f/stop
advantage.

See:

http://www.pcworld.com/article/id,13...s/article.html

David


If it works as described, it just might fix my single complaint about
my wife's camera, poor low light operation. We shall see.

Ah, but it does not fix your wife's camera. My wife noticed right away
that this was an opportunity to spend money, not a solution for my
biggest complaint about my camera. :-)

Paul Allen

[A complimentary Cc of this posting was NOT [per weedlist] sent to
Kennedy McEwen
], who wrote in article :
WBWG
BWGW
WGWR
GWRW


You need a complete 4x4 array of pixels to guarantee you get all of the
colour information, compared to a 2x2 pixel array for the Bayer matrix.
As you point out, this produces extremely poor colour resolution, as

Nonsense. You just have a wrong mental picture. Redraw what you
wrote shifted 1 horizontally:

BWGW
WGWB
GWBW
WBWG

Any 2x2 square contains all pixels. (Of course, this is not what *I*
consider optimal: white/cyan/yellow, as in

WC
YW

repeated.)

However, it doesn't have to be that way, which is why I suspect a PC
World graphic artist screw up.

Me too.

Hope this helps,
Ilya

In article , Ilya Zakharevich
writes
[A complimentary Cc of this posting was NOT [per weedlist] sent to
Kennedy McEwen
], who wrote in article
:
WBWG
BWGW
WGWR
GWRW


You need a complete 4x4 array of pixels to guarantee you get all of the
colour information, compared to a 2x2 pixel array for the Bayer matrix.
As you point out, this produces extremely poor colour resolution, as

Nonsense. You just have a wrong mental picture. Redraw what you
wrote shifted 1 horizontally:

BWGW
WGWB
GWBW
WBWG

Well that isn't what I wrote above shifted one pixel left. For a start,
it doesn't have any red! This is the array in the PC World diagram
shifted one pixel left from what I wrote above, with the leftmost column
wrapping around to the right:
BWGW
WGWB
GWRW
WRWG


Any 2x2 square contains all pixels.

Really? Look at the first 2x2 set:
BW
WG
That contains all the colours? Doh! Where did the red go?

And the last 2x2 set:
RW
WG
Oops, no blue!

Of course. you can "estimate" that blue (or red in the example above)
from the difference between the two whites and the red and green pixels,
but it doesn't work very well. Just calculate the noise on that
resulting blue (or red) sample! Just from shot noise alone it is more
than 2x worse than either the red or green pixel, and readout noise
makes things worse. Then you have the situation when the white
saturates and the missing colour simply cannot be estimated by
difference filtering at all.

No, you need 4x4 to get the full colour information unambiguously and,
as Roger pointed out, that is a lot lower colour resolution than the
standard Bayer array produces.

Of course, this is not what *I*
consider optimal: white/cyan/yellow, as in

WC
YW

repeated.

Subtractive filtering doesn't seem to have become too popular on
sensors. It has been tried, but has always been found wanting for one
reason or another compared to the simple Bayer matrix, mainly due to
similar reasons as above.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's ****ed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)