Digital Cameras: Does Pixel Size Matter?

Hi,
I've quantified digital camera sensor performance, comparing
photon-noise-limited DSLRs and point and shoot cameras.
Good cameras are now photon noise limited and this sets
basic properties that depend on pixel size and how many photons
can be collected by the sensor. The photon limits and resulting
gain factors explain the major differences observed between the
large pixels of DSLRs and the small pixels found in point and
shoot cameras.

Article at:
http://clarkvision.com/imagedetail/d...el.size.matter

Roger

"Roger N. Clark (change username to rnclark)" wrote:

http://clarkvision.com/imagedetail/d...el.size.matter

Thanks! It's great to have this stuff done so well and be on a page we can
reference.

David J. Littleboy
Tokyo, Japan

You do, however, still have to match your expected image scale to your pixel
size, hence the paucity of wider angle lenses for CCD sensors (Nyquist
criteria, etc, etc...).
________________
G. Paleologopoulos


"Roger N. Clark (change username to rnclark)" wrote in
message ...

Hi,
I've quantified digital camera sensor performance, comparing
photon-noise-limited DSLRs and point and shoot cameras.
Good cameras are now photon noise limited and this sets
basic properties that depend on pixel size and how many photons
can be collected by the sensor. The photon limits and resulting
gain factors explain the major differences observed between the
large pixels of DSLRs and the small pixels found in point and
shoot cameras.

Article at:
http://clarkvision.com/imagedetail/d...el.size.matter

Roger

Roger N. Clark (change username to rnclark) wrote:
Hi,
I've quantified digital camera sensor performance, comparing
photon-noise-limited DSLRs and point and shoot cameras.
Good cameras are now photon noise limited and this sets
basic properties that depend on pixel size and how many photons
can be collected by the sensor. The photon limits and resulting
gain factors explain the major differences observed between the
large pixels of DSLRs and the small pixels found in point and
shoot cameras.

Article at:
http://clarkvision.com/imagedetail/d...el.size.matter

Roger

Roger,

Many thanks for another stimulating view.

There are some typos:

- 1st paragraph: "dod not get"

- table 4 - shouldn't the S60 shadow total be 332?

Questions:

1 - ISO measurement

What it all says to me is that the ISO figures are grossly misleading. If
all cameras were required to give the same SNR at (say) 18% grey level,
then couldn't we just say that the S60 had ISO 100 sensitivity and the 10D
ISO 400 sensitivity? What are we doing wrong in stating ISO sensitivities
today?

2 - Number of pixels for a given sensor size.

I know I've raised this before but I'll do so again. If you have a given
sensor (let's think small sensor here - e.g. 8.8 x 6.6mm), then you could,
for example, have either 2MPixels at one size or 8MPixels at half the
linear size or a quarter of the area. Neglecting packing efficiency, you
could have a higher-resolution image with poorer SNR in each pixel, or a
lower resolution image with higher SNR per pixel. Which does the eye
prefer in typical viewing conditions? The image with the smaller pixels
will appear crisper. How is the different noise spectrum perceived by the
eye?

Cheers,
David

David J Taylor wrote:
Roger N. Clark (change username to rnclark) wrote:
Article at:
http://clarkvision.com/imagedetail/d...el.size.matter


Roger,

Many thanks for another stimulating view.

There are some typos:

- 1st paragraph: "dod not get"

- table 4 - shouldn't the S60 shadow total be 332?

Thanks David,
You are correct. I fixed these.

Questions:

1 - ISO measurement

What it all says to me is that the ISO figures are grossly misleading. If
all cameras were required to give the same SNR at (say) 18% grey level,
then couldn't we just say that the S60 had ISO 100 sensitivity and the 10D
ISO 400 sensitivity? What are we doing wrong in stating ISO sensitivities
today?

This is a good question. I think the answer is partly that
ISO was originally defined as a density on film at a given
exposure time for a given light source intensity. If one takes
a percentage of full well capacity in electronic sensors,
as a substitute for density on film, then we see ISO definition
has to do with how many photons are collected as a percentage
of the total possible. That says nothing about noise, just as
with film the specification says nothing about noise
(or film grain).

Side note:
Because film has a much lower quantum efficiency than CCDs or
CMOS sensors, it collects many fewer photons, about 6 times
lower (at least). Scaling the numbers here,
we could compute the number of photons incident per
square micron at a given ISO, and f/ratio. It appears to me
that film too may be close to photon noise limited, or at
least a large part of the noise we see in film, which we call
grain is due to photon statistics, added to the random
distribution of film grains, which induces its own noise.
But I believe the photon noise is the larger of the two
components.

2 - Number of pixels for a given sensor size.

I know I've raised this before but I'll do so again. If you have a given
sensor (let's think small sensor here - e.g. 8.8 x 6.6mm), then you could,
for example, have either 2MPixels at one size or 8MPixels at half the
linear size or a quarter of the area. Neglecting packing efficiency, you
could have a higher-resolution image with poorer SNR in each pixel, or a
lower resolution image with higher SNR per pixel. Which does the eye
prefer in typical viewing conditions? The image with the smaller pixels
will appear crisper. How is the different noise spectrum perceived by the
eye?

This is a very good question, and I have been researching this
subject. It appears that to first order there is an even trade for
signal-to-noise and spatial resolution. So in your example,
the images would appear similar. But because it is
subjective, some people will prefer one, others the other,
but they would be judged close, especially compared to the
same images with equal signal-to-noise ratios. I've created
images like this and that is what I see and what others have
indicated. But it is a subjective test. I plan on creating a
series of images, trading spatial resolution and signal-to-noise
ratio and presenting that on a web page for people to judge for
themselves. But it will be a few months before I get this done.

Roger

g n p wrote:

You do, however, still have to match your expected image scale to your pixel
size, hence the paucity of wider angle lenses for CCD sensors (Nyquist
criteria, etc, etc...).
________________
G. Paleologopoulos

I don't see what this has to do with sensor noise versus
pixel size, which is all my page addresses.
Could you please clarify what you mean?

Roger


"Roger N. Clark (change username to rnclark)" wrote in
message ...

Hi,
I've quantified digital camera sensor performance, comparing
photon-noise-limited DSLRs and point and shoot cameras.
Good cameras are now photon noise limited and this sets
basic properties that depend on pixel size and how many photons
can be collected by the sensor. The photon limits and resulting
gain factors explain the major differences observed between the
large pixels of DSLRs and the small pixels found in point and
shoot cameras.

Article at:
http://clarkvision.com/imagedetail/d...el.size.matter

Roger

Roger N. Clark (change username to rnclark) wrote:
[]
This is a good question. I think the answer is partly that
ISO was originally defined as a density on film at a given
exposure time for a given light source intensity. If one takes
a percentage of full well capacity in electronic sensors,
as a substitute for density on film, then we see ISO definition
has to do with how many photons are collected as a percentage
of the total possible. That says nothing about noise, just as
with film the specification says nothing about noise
(or film grain).

Side note:
Because film has a much lower quantum efficiency than CCDs or
CMOS sensors, it collects many fewer photons, about 6 times
lower (at least). Scaling the numbers here,
we could compute the number of photons incident per
square micron at a given ISO, and f/ratio. It appears to me
that film too may be close to photon noise limited, or at
least a large part of the noise we see in film, which we call
grain is due to photon statistics, added to the random
distribution of film grains, which induces its own noise.
But I believe the photon noise is the larger of the two
components.

I would welcome a more sensible definition of ISO as applied to digital
cameras. Percentage well filling without specifying a noise level is a
plain nonsense. Perhaps then we could move on from the DLSR versus Point
& Shoot argument and simply say if you can be satisfied with a low ISO
camera go P&S and if you need high-ISO go DLSR?


2 - Number of pixels for a given sensor size.

I know I've raised this before but I'll do so again. If you have a
given sensor (let's think small sensor here - e.g. 8.8 x 6.6mm),
then you could, for example, have either 2MPixels at one size or
8MPixels at half the linear size or a quarter of the area. Neglecting
packing efficiency, you could have a higher-resolution
image with poorer SNR in each pixel, or a lower resolution image
with higher SNR per pixel. Which does the eye prefer in typical
viewing conditions? The image with the smaller pixels will appear
crisper. How is the different noise spectrum perceived by the eye?

This is a very good question, and I have been researching this
subject. It appears that to first order there is an even trade for
signal-to-noise and spatial resolution. So in your example,
the images would appear similar. But because it is
subjective, some people will prefer one, others the other,
but they would be judged close, especially compared to the
same images with equal signal-to-noise ratios. I've created
images like this and that is what I see and what others have
indicated. But it is a subjective test. I plan on creating a
series of images, trading spatial resolution and signal-to-noise
ratio and presenting that on a web page for people to judge for
themselves. But it will be a few months before I get this done.

Roger

When I was involved with this professionally some years ago, we used a
weighted SNR based on the sensitivity of the eye & brain at various
spatial frequencies. Rather as in audio where there are various weighting
curves you can use to make the measured SNR tie up better with the
subjective SNR than an unweighted flat frequency response figure. This
attempts to remove at least some of the subjective element, but you are
right that different observers will have different preferences making a
weighted SNR only representative of a median of observers. Viewing
conditions would need to be standardised as well!

Cheers,
David

David J Taylor wrote:

I would welcome a more sensible definition of ISO as applied to digital
cameras. Percentage well filling without specifying a noise level is a
plain nonsense. Perhaps then we could move on from the DLSR versus Point
& Shoot argument and simply say if you can be satisfied with a low ISO
camera go P&S and if you need high-ISO go DLSR?

I would like to see the manufacturers specify the actual sensor
size instead of the "type." How many people really know
what a 4/3" sensor is?

Then if the manufacturers would specify the full well capacity
and read noise, consumers would learn quickly that you want
full well high, and read noise low, and people would have the
information needed to make an informed decision.

Roger

Roger N. Clark (change username to rnclark) wrote:
David J Taylor wrote:

I would welcome a more sensible definition of ISO as applied to
digital cameras. Percentage well filling without specifying a noise
level is a plain nonsense. Perhaps then we could move on from the
DLSR versus Point & Shoot argument and simply say if you can be
satisfied with a low ISO camera go P&S and if you need high-ISO go
DLSR?

I would like to see the manufacturers specify the actual sensor
size instead of the "type." How many people really know
what a 4/3" sensor is?

Then if the manufacturers would specify the full well capacity
and read noise, consumers would learn quickly that you want
full well high, and read noise low, and people would have the
information needed to make an informed decision.

Roger

I agree that you want that in the specification somewhere, but those
aren't convenient measures for everyday use and, to be honest, is the
actual sensor size as its effect on picture quality. I was thinking more
like this: if picture quality was rated on a scale of 1 - 10 (say) and
actually meaning the weighted SNR, you could make comparisons like:

Picture quality 8:
DLSR has ISO 400
P&S 1 has ISO 80
P&S 2 cannot achieve PQ 8 at full resolution

Picture quality 5:
DLSR has ISO 1600
P&S 1 has ISO 200
P&S 2 has ISO 100

Cheers,
David

Your analysis was excellent.
My comment was sort of an aside, a footnote to your page, to not mislead the
reader into thinking that large pixels are a general cure-all, and that a
balance must be found between spatial resolution and noise.
Stuff we into CCD astrophotography have beaten to death many, many years
ago!
________________
G. Paleologopoulos



"Roger N. Clark (change username to rnclark)" wrote in
message ...

g n p wrote:

You do, however, still have to match your expected image scale to your
pixel size, hence the paucity of wider angle lenses for CCD sensors
(Nyquist criteria, etc, etc...).
________________
G. Paleologopoulos

I don't see what this has to do with sensor noise versus
pixel size, which is all my page addresses.
Could you please clarify what you mean?

Roger