Image Stabilization vs Noise

jpc wrote:
On Thu, 21 Dec 2006 10:44:48 -0800, Phil Wheeler
wrote:

jpc wrote:
Based on some published information from a review site, IS point and
shoot camera seem to be more noisy than non IS camera. This I'm
blaiming on electrical pickup from the piezo motors used to move
either a lens element or the sensor.

Anyone have any experience or comments on this topic


It would help if you provided a link to the review
site.


Yesterday was a busy day, so I didn't get a change to get back to
anyone's post.

The review site I'm using is imaging-resource.com. While I believe
they mentioned high noise in a review of a Canon IS P&S, what I've
done is use the DaveBox images to do my own noise analysis. If you
take the trouble to extract the information, these images are a
treasure trove of information on camera performance since the
imaging-resource people have been taking pictures of the same target
under the same lighting conditions for the last 12 years.

Here is my procedure if you want to duplicate the data I have on my
screen right now.

Go to the Canon 5D, Canon S3IS and Oly SP350 reviews. Download the
200 and 400 ISO low light (11lumen) davebox images of all three
cameras. These are at the end of the review and are the one's I've
been using since 11 lumens is roughly the light level you'd see at
night on a lighed city street.

Next Google ImageJ and go to the NIH website for a free download.
ImageJ is an excellent image analysis package that will do many things
but the only thing you have to do is hit the line icon on the tool bar
and drag a line down the grey scale section on the right central part
of your davebox images. Then hit Cnrl-K and a staircase graph will pop
up. You'll see the value (0-255) for each step in the grey scale with
the noise superimposed on the steps.

The Canon5D is our standard. Notice how all the noise is low and all
the steps are easily seen. (The bumpiness near the bottom is a target
problem). Also notice how the noise stays constant.

Next look at at SP350 graph--one of the tiny pixel (5 square microns)
camera that many in the news group like to trash. It's noise is 3 time
worse than the 5D, which is exactly what you'd expect since the sensor
area is 9 times smaller. And like the 5D the noise is pretty much
constant as you move down the graph

Now look at the S3IS graph. Not only is the noise not even close to
being constant, it's over 10 time worse than the 5D in the dark area
of the grey scale and 3-4 times worse that the SP350.

Note--I'd argue that since ISO numbers are just gain settings to
compare the sensor noise in cameras, you should start at lowest and X1
gain-- no matter what the marketing folks decided to call that setting
-- and count up gain steps. So my numbers are from comparing the 200
ISO graphs of the 5D and Sp350 (both cameras start at ISO50) with the
400 ISO of the S3IS. If you disagree and compare graphs labeled with
the same ISO numbers, the results are the same, just a little less
obvious.

So what going on? Since Canon does know how to make good low
noise cameras, I'm guessing the problem is caused by electrical pickup
from the piezo motors. With the sensor ouputs measured in microvolts
and the piezo motors being hit continiously at much higher voltages
the design must be a noise-engineer's nightmare.

What you're seeing has absolutely nothing whatever to do with IS.
It has to do with the sensor size, type, and the signal amplificatino
required to eek data out of much smaller sensor points. The full frame on
the 5D has comparatively HUGE points, which are naturally capable of
gathering more light...meaning less need to amplify the signal.

This is one of many reasons why small sensor point-and-shoots can't compete
with DSLRs in terms of noise adn high ISO performance. Again... NOT an IS
issue.


I have three IS cameras and four IS lenses for my
DSLR. None of these types of noise has been an issue.


I wouldn't expect to see this on a DSLR using IS lens since the
piezo motors are much farther away from the sensor electronics.


Any comments?

jpc

--
Images (Plus Snaps & Grabs) by Mark² at:
www.pbase.com/markuson

Skip wrote:
"jpc" wrote in message
...
Based on some published information from a review site, IS point and
shoot camera seem to be more noisy than non IS camera. This I'm
blaiming on electrical pickup from the piezo motors used to move
either a lens element or the sensor.

Anyone have any experience or comments on this topic

jpc

I'm guessing it's because the p&s IS cameras have "digital"
stabilization, which, in many cases, is merely a bump up for ISO and
shutter speed, which, of course, results in more noise...

If he's talking about non-optical IS, then that's a different story
altogether, and has little or no relation to optical IS lenses attached to
DSLRs...

--
Images (Plus Snaps & Grabs) by Mark² at:
www.pbase.com/markuson

On 22 Dec 2006 12:46:57 -0800, DougL wrote:

We spend several days and a lot of words trying to establish exactly
who has what. As far as the thread participants were concerned, we
ended up concluding that point and shoot cameras did NOT have OIS. Can
you supply a pointer that says otherwise?

Would be interesting to know who really does image stabilization
(either CCD-shift or optical) in a point-and-shoot.

Yes, "moving mirror" was my shorthand. "Moving optic" would have been
more accurate.

After a brief search, the cheapest camera I could find with a real
optical stabilizer (using a lens-shift system) is the Panasonic LZ3,
$160 at B&H.

Plenty of other P&Ss, from a whole slew of makers, use real mechanical
stabilizers.

-dms

On 22 Dec 2006 12:53:21 -0800, DougL wrote:

As I said, these inexpensive cameras just have DIS, and there isn't
anything moving in this "stabiliztion system". Not even electrons
between pixels.

Go to http://www.dpreview.com/reviews/panasoniclz2/page5.asp
and look at the sample images on that page. It's a real
lens-shift stabilizer.

This year's version of the same camera goes for about $150ish, which
definitely puts it in the 'inexpensive' category.

Never heard of moving the sensor. As in moving the CCD?
Engineering-wise, that would be pretty challenging.

The stabilizer system that Sony inherited from Minolta is a sensor-shift
system, as is the stabilizer that Pentax uses in their DSLRs.

-dms

On Sat, 23 Dec 2006 01:50:14 -0800, "Mark²" mjmorgan(lowest even
number wrote:

jpc wrote:
On Thu, 21 Dec 2006 10:44:48 -0800, Phil Wheeler
wrote:

jpc wrote:
Based on some published information from a review site, IS point and
shoot camera seem to be more noisy than non IS camera. This I'm
blaiming on electrical pickup from the piezo motors used to move
either a lens element or the sensor.

Anyone have any experience or comments on this topic


It would help if you provided a link to the review
site.


Yesterday was a busy day, so I didn't get a change to get back to
anyone's post.

The review site I'm using is imaging-resource.com. While I believe
they mentioned high noise in a review of a Canon IS P&S, what I've
done is use the DaveBox images to do my own noise analysis. If you
take the trouble to extract the information, these images are a
treasure trove of information on camera performance since the
imaging-resource people have been taking pictures of the same target
under the same lighting conditions for the last 12 years.

Here is my procedure if you want to duplicate the data I have on my
screen right now.

Go to the Canon 5D, Canon S3IS and Oly SP350 reviews. Download the
200 and 400 ISO low light (11lumen) davebox images of all three
cameras. These are at the end of the review and are the one's I've
been using since 11 lumens is roughly the light level you'd see at
night on a lighed city street.

Next Google ImageJ and go to the NIH website for a free download.
ImageJ is an excellent image analysis package that will do many things
but the only thing you have to do is hit the line icon on the tool bar
and drag a line down the grey scale section on the right central part
of your davebox images. Then hit Cnrl-K and a staircase graph will pop
up. You'll see the value (0-255) for each step in the grey scale with
the noise superimposed on the steps.

The Canon5D is our standard. Notice how all the noise is low and all
the steps are easily seen. (The bumpiness near the bottom is a target
problem). Also notice how the noise stays constant.

Next look at at SP350 graph--one of the tiny pixel (5 square microns)
camera that many in the news group like to trash. It's noise is 3 time
worse than the 5D, which is exactly what you'd expect since the sensor
area is 9 times smaller. And like the 5D the noise is pretty much
constant as you move down the graph

Now look at the S3IS graph. Not only is the noise not even close to
being constant, it's over 10 time worse than the 5D in the dark area
of the grey scale and 3-4 times worse that the SP350.

Note--I'd argue that since ISO numbers are just gain settings to
compare the sensor noise in cameras, you should start at lowest and X1
gain-- no matter what the marketing folks decided to call that setting
-- and count up gain steps. So my numbers are from comparing the 200
ISO graphs of the 5D and Sp350 (both cameras start at ISO50) with the
400 ISO of the S3IS. If you disagree and compare graphs labeled with
the same ISO numbers, the results are the same, just a little less
obvious.

So what going on? Since Canon does know how to make good low
noise cameras, I'm guessing the problem is caused by electrical pickup
from the piezo motors. With the sensor ouputs measured in microvolts
and the piezo motors being hit continiously at much higher voltages
the design must be a noise-engineer's nightmare.

What you're seeing has absolutely nothing whatever to do with IS.
It has to do with the sensor size, type, and the signal amplificatino
required to eek data out of much smaller sensor points. The full frame on
the 5D has comparatively HUGE points, which are naturally capable of
gathering more light...meaning less need to amplify the signal.

As I stated in my follow-up post to David Taylor the 5D has a pixel
size of ~70 sq microns and an active sensor size of ~45 sq microns.
The Oly SP350--a much underrated camerra I ended up buying--has an
active sensor size of ~5 sq microns.

Photon shot noise, Poison noise, or the random way Momma Nature tosses
light into your lens is "roughly" determined by pixel size. The real
parameter is pixel well depth--something the camera manufactures have
decided not to tell us about.

Since Momma Nature noise is determined by the sq root of the number of
photo electrons collected and the square root of 9 is 3, the SP350
has a S/N 3 time worse than the 5D in the regions where Momma Nature
noise dominates.

This is what theory predicts, this is what the data from Imaging-
resourse shows and most important to me--this is how the SP350 I
bought performed when I tested it.

So I've lost a stop and a half of signal to noise by buying a camera
that some in this group automatically trash .But since I paid 20X less
that a 5D with lens the price /performance ratio is definately in my
favor.

And of couse, I can carry around my camera outfit--camera, batteries,
cards, telephoto and wide angle extender lens, even a mini tripod-- in
my coat pocket. That's something you can not say about a 5D and its
collection of lens.


This is one of many reasons why small sensor point-and-shoots can't compete
with DSLRs in terms of noise adn high ISO performance. Again... NOT an IS
issue.

The Imaging-resouce noise data for the S3IS does NOT follow the
theory and physics of Momma Nature noise. The noise in the dark
patches--the shadow areas in low light photography where S/N is most
important-- is 4 X worse than it should be.

From that I can draw two conclusions. The first is that image
stabization in a high end P&S is not the solution to all problems like
the camera sellers want you to believe when you plop down your $400.
The second is that the S3IS is overall a lousy camera.

So if anyone who own a IS P&S is willing to spend a couple hours of
their life to help answer this burning question, I'm willing to offer
advice and comentary. If I'm wrong, I'm wrong. But if I'm right, 2
bucks worth of chain vs 400 bucks worth of camera makes for an
excelent price/perfomance ratio


jpc

"jpc" wrote:

As I stated in my follow-up post to David Taylor the 5D has a pixel
size of ~70 sq microns and an active sensor size of ~45 sq microns.
The Oly SP350--a much underrated camerra I ended up buying--has an
active sensor size of ~5 sq microns.

Photon shot noise, Poison noise, or the random way Momma Nature tosses
light into your lens is "roughly" determined by pixel size. The real
parameter is pixel well depth--something the camera manufactures have
decided not to tell us about.

Since Momma Nature noise is determined by the sq root of the number of
photo electrons collected and the square root of 9 is 3, the SP350
has a S/N 3 time worse than the 5D in the regions where Momma Nature
noise dominates.

This is what theory predicts, this is what the data from Imaging-
resourse shows and most important to me--this is how the SP350 I
bought performed when I tested it.

So far so good, but...

So I've lost a stop and a half of signal to noise by buying a camera
that some in this group automatically trash .But since I paid 20X less
that a 5D with lens the price /performance ratio is definately in my
favor.

Oops. No. SNR vs. ISO stops don't work that way. An exposure at ISO 100
captures twice as many photons as ISO 200, so cutting the sensor area in
half means your ISO 100 has the same SNR ratio as the larger pixel's ISO
200. (That is, a one stop higher ISO has 1.414 times the noise, not twice
the noise.)

(The price/performance ratio is still in your favor, thoughg.)

So you've lost _more than three_ stops of SNR ratio. The 5D at ISO 800 looks
better than the P&S at ISO 100. And the 5D user has f/1.4 and f/2.0 lenses
if s/he wants, which are non-existent and rare in the P&S world nowadays.

This is one of many reasons why small sensor point-and-shoots can't
compete
with DSLRs in terms of noise adn high ISO performance. Again... NOT an
IS
issue.

The Imaging-resouce noise data for the S3IS does NOT follow the
theory and physics of Momma Nature noise. The noise in the dark
patches--the shadow areas in low light photography where S/N is most
important-- is 4 X worse than it should be.

I don't know if this is wrong (due to the above mistake) or not, but it
might be.

David J. Littleboy
Tokyo, Japan

On Sat, 23 Dec 2006 01:46:34 -0800, "Mark²" mjmorgan(lowest even
number wrote:

jpc wrote:
Based on some published information from a review site, IS point and
shoot camera seem to be more noisy than non IS camera. This I'm
blaiming on electrical pickup from the piezo motors used to move
either a lens element or the sensor.

Anyone have any experience or comments on this topic


Comments? Yes:
Complete ********.

See my reply above. I'm talking physics, data, and proceedures for
evaluating the data if anyone wishes to confirm what I'm saying.

I'm even willing to cheerfully admit I'm wrong if anyone can come up
with an alternate and better explanation to the conclusions I've drawn
from the data.

What more can you ask.

jpc

On Sat, 23 Dec 2006 23:44:29 +0900, "David J. Littleboy"
wrote:


"jpc" wrote:

As I stated in my follow-up post to David Taylor the 5D has a pixel
size of ~70 sq microns and an active sensor size of ~45 sq microns.
The Oly SP350--a much underrated camerra I ended up buying--has an
active sensor size of ~5 sq microns.

Photon shot noise, Poison noise, or the random way Momma Nature tosses
light into your lens is "roughly" determined by pixel size. The real
parameter is pixel well depth--something the camera manufactures have
decided not to tell us about.

Since Momma Nature noise is determined by the sq root of the number of
photo electrons collected and the square root of 9 is 3, the SP350
has a S/N 3 time worse than the 5D in the regions where Momma Nature
noise dominates.

This is what theory predicts, this is what the data from Imaging-
resourse shows and most important to me--this is how the SP350 I
bought performed when I tested it.

So far so good, but...

So I've lost a stop and a half of signal to noise by buying a camera
that some in this group automatically trash .But since I paid 20X less
that a 5D with lens the price /performance ratio is definately in my
favor.

Oops. No. SNR vs. ISO stops don't work that way. An exposure at ISO 100
captures twice as many photons as ISO 200, so cutting the sensor area in
half means your ISO 100 has the same SNR ratio as the larger pixel's ISO
200. (That is, a one stop higher ISO has 1.414 times the noise, not twice
the noise.)

Agreed. Half the photoelectrons in the pixel well, 1.4 times more
noise

(The price/performance ratio is still in your favor, thoughg.)

So you've lost _more than three_ stops of SNR ratio. The 5D at ISO 800 looks
better than the P&S at ISO 100.

Right,again. My mental miscalculation

ISO 400 (2 stops and 4X gain) gives 2X noise. ISO 800 (3 stops and 8X
gain) give 3X noise. Which is what the imaging-resource data says--a
3X noise difference between the two camera.

And the 5D user has f/1.4 and f/2.0 lenses
if s/he wants, which are non-existent and rare in the P&S world
nowadays. This is one of many reasons why small sensor
point-and-shoots can't compete with DSLRs in terms of noise adn high
ISO performance. Again... NOT an IS issue.

Agreed-partially. If we're talking about photon noise--the noise
source where sensor size is an issue--it doesn't matter is you fill
the pixel well in a 1/1000 of a second or a 1/10 off a second. The S/N
is still the same until you reach a shutter speed where thermal noise
becomes a problem.

A f1.4 lens obviously will fill the pixel well in a 1/4 the time as a
f2.8 lens under identical conditions. So my buddy's $5000 colection of
camera and lens will beat my $225 camera, it's heavy weigh vs light
weigh, but I don't think lens speeds have much to do with my question.


The Imaging-resouce noise data for the S3IS does NOT follow the
theory and physics of Momma Nature noise. The noise in the dark
patches--the shadow areas in low light photography where S/N is most
important-- is 4 X worse than it should be.

I don't know if this is wrong (due to the above mistake) or not, but it
might be.

Here's my question restated. The SP350 and S3IS have roughly similar
size pixel and should have roughly 3-4X more photon noise than the 5D.
The S3!S has 10-12X more noise.

Where is the extra noise coming from?

jpc

"DougL" wrote in message
ps.com...

Skip wrote:
"J. Clarke" wrote in message
...
On Thu, 21 Dec 2006 14:43:06 -0800, DougL wrote:

Skip wrote:
"jpc" wrote in message
...
Based on some published information from a review site, IS point
and
shoot camera seem to be more noisy than non IS camera. This I'm
blaiming on electrical pickup from the piezo motors used to move
either a lens element or the sensor.

Anyone have any experience or comments on this topic

jpc

I'm guessing it's because the p&s IS cameras have "digital"
stabilization,
which, in many cases, is merely a bump up for ISO and shutter speed,
which,
of course, results in more noise...

--
Skip Middleton
www.shadowcatcherimagery.com
www.pbase.com/skipm

Exactly right, as finally concluded in this long thread over the last
two days.

http://tinyurl.com/yj7mpq

Turn it off, if you can, if you don't need it.

Note too that digital image stabilization (DIS) has neither a
stabilization sensor, nor does any image motion. Just ISO bump up,
which amplifies system noise as well. We finally concluded that
calling
this IS was quite misrepresentative.

Optical image stabilization (OIS) has a tilt sensor, and a moving
mirror to shift the image.

Charge motion image stabilization has a tilt sensor, and shifts the
image digitally in the pixels electronically.

Inexpensive consumer cameras just have DIS. No sensor, no piezo
motors,
no nothing.

Oh, GAWD.

First most point and shoot cameras with image stabilization have
optical
image stabilization just like Canon and Nikon DSLRs. It's only the
Fuji
and possibly some other other cheap POS cameras that pretend that they
do by increasing the ISO. Second, optical image stabilization does not
use a "moving mirror", one of the elements in the lens moves, those
elements all being lenses unless its a catadioptric.

--
--John
to email, dial "usenet" and validate
(was jclarke at eye bee em dot net)

The Oly FE-170 I was flummoxed into buying for my daughter for Christmas
has
"digital image stabilization," not an expensive camera, at $150, but from
a
respected mfr.
I won't argue your other points, (moving mirror?), but some of them move
the
sensor, don't they?

--
Skip Middleton
www.shadowcatcherimagery.com
www.pbase.com/skipm

As I said, these inexpensive cameras just have DIS, and there isn't
anything moving in this "stabiliztion system". Not even electrons
between pixels.

Never heard of moving the sensor. As in moving the CCD?
Engineering-wise, that would be pretty challenging.

Several DSLRs have that feature, now, the Sony A100, for one.

--
Skip Middleton
www.shadowcatcherimagery.com
www.pbase.com/skipm

jpc wrote in message ...
On 22 Dec 2006 12:53:21 -0800, "DougL" wrote:



Never heard of moving the sensor. As in moving the CCD?
Engineering-wise, that would be pretty challenging.

Agree with that 100 %.

Here's my version of camera shake correction---AKA IS

Take a five foot length of light chain, attach it to a 1/4-20 bolt
and screw into the tripod mounting hole of the camera. Wrap the camera
strap around your neck and arm like it is a rifle sling. Frame your
shot, step on the chain, and pull the camera taunt.

Instant tripod.

After some practice, I been able to take some reasonable hand held
pictures with 1/4 to 1/2 exposure times. Don't know how this stacks up
against the fancier version but I bet it's in the ball park.

jpc

I've carried around a length of light nylon twine with a bolt for 20 years,
works well in a pinch, but not as well as IS or a tripod.

--
Skip Middleton
www.shadowcatcherimagery.com
www.pbase.com/skipm