Sign in with
Sign up | Sign in
Your question

determining native iso of sensor

Last response: in Digital Camera
Share
Anonymous
February 13, 2005 11:51:10 AM

Archived from groups: rec.photo.digital.slr-systems (More info?)

Is there a way to determine the native ISO of the sensor in my Canon
20D? From what I've heard, it isn't necessarily the lowest ISO
setting. Someone said the sensor in the 10D was best at 200 and that
100 was actually a slight interpolation.
February 13, 2005 5:42:46 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

On Sun, 13 Feb 2005 08:51:10 -0800, drs@canby.com wrote:

>Is there a way to determine the native ISO of the sensor in my Canon
>20D? From what I've heard, it isn't necessarily the lowest ISO
>setting. Someone said the sensor in the 10D was best at 200 and that
>100 was actually a slight interpolation.


That's a bit like asking "what is the native volume setting on my stereo?"

ISO in a digital camera is just the calibration of the system as compared with
film... there is no 'native' ISO.

For example, if an ISO 100 film has a perfect gray scale image with a certain
light and an F stop and a shutter speed and lens... then that same photo when
taken with a digital camera (and all the same settings) will mean the camera is
set to ISO 100... in other words, the engineer will adjust the gain of the
prototype camera to get the same picture and then gain is said to be set to ISO
100...

Normally the lowest ISO number means the best signal to noise ratio and the best
picture, but I imagine that a camera could be rated to be better at a higher
number, due to various compromises, you need to know exactly why someone said
200 is better...
Anonymous
February 13, 2005 7:20:12 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

drs@canby.com wrote:

> Is there a way to determine the native ISO of the sensor in my Canon
> 20D? From what I've heard, it isn't necessarily the lowest ISO
> setting. Someone said the sensor in the 10D was best at 200 and that
> 100 was actually a slight interpolation.

Native? It's a sensor that 'accumulates' a 'charge' of light over time.

For example, it does not care if the exposure time is 1/4000 or 40 seconds. At
some point there isn't enough light ("zero" accumlation) and at some point there
is too much (saturated -> burt highlights).

What you might be referring to is "at what ISO setting is the gain set to 1" (0
dB). Even then I doubt the answer is that simple.

Best ISO? I believe if you go to dpreview and look at the noise graphs you'll
find the D20 is least noisy at ISO 100.
http://www.dpreview.com/reviews/konicaminolta7d/page18....

Cheers,
Alan

--
-- r.p.e.35mm user resource: http://www.aliasimages.com/rpe35mmur.htm
-- r.p.d.slr-systems: http://www.aliasimages.com/rpdslrsysur.htm
-- [SI] gallery & rulz: http://www.pbase.com/shootin
-- e-meil: there's no such thing as a FreeLunch.
Related resources
Anonymous
February 13, 2005 7:20:13 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

Alan Browne wrote:
> drs@canby.com wrote:
>
>> Is there a way to determine the native ISO of the sensor in my Canon
>> 20D? From what I've heard, it isn't necessarily the lowest ISO
>> setting. Someone said the sensor in the 10D was best at 200 and that
>> 100 was actually a slight interpolation.
>
>
> Native? It's a sensor that 'accumulates' a 'charge' of light over time.
>
> For example, it does not care if the exposure time is 1/4000 or 40
> seconds. At some point there isn't enough light ("zero" accumlation)
> and at some point there is too much (saturated -> burt highlights).
>
> What you might be referring to is "at what ISO setting is the gain set
> to 1" (0 dB). Even then I doubt the answer is that simple.
>
> Best ISO? I believe if you go to dpreview and look at the noise graphs
> you'll find the D20 is least noisy at ISO 100.
> http://www.dpreview.com/reviews/konicaminolta7d/page18....

Then using the term "native" in this context is quite clear to most. See
JP Sheehy's reply also.
--
John McWilliams
Anonymous
February 13, 2005 8:44:24 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

drs@canby.com wrote in news:361v01dld4tp5d67omla3lbck4j3n6ajgo@4ax.com:

> Is there a way to determine the native ISO of the sensor in my Canon
> 20D? From what I've heard, it isn't necessarily the lowest ISO
> setting. Someone said the sensor in the 10D was best at 200 and that
> 100 was actually a slight interpolation.

As I understand it - the ISO number of a CCD sensor is determined
by the white clipping point, just as it is done for slide film.

For low contrast subjects you can then use a lower ISO number.


/Roland
Anonymous
February 14, 2005 12:01:51 AM

Archived from groups: rec.photo.digital.slr-systems (More info?)

In message <Xns95FCBEA20C2F5klotjohan@130.133.1.4>,
Roland Karlsson <roland_dot_karlsson@bonetmail.com> wrote:

>drs@canby.com wrote in news:361v01dld4tp5d67omla3lbck4j3n6ajgo@4ax.com:
>
>> Is there a way to determine the native ISO of the sensor in my Canon
>> 20D? From what I've heard, it isn't necessarily the lowest ISO
>> setting. Someone said the sensor in the 10D was best at 200 and that
>> 100 was actually a slight interpolation.
>
>As I understand it - the ISO number of a CCD sensor is determined
>by the white clipping point, just as it is done for slide film.
>
>For low contrast subjects you can then use a lower ISO number.

I think that drs' question is actually something like, "is the 20D's ISO
100 actually its ISO 200, pulled one stop", implying a loss of one stop
of highlights in the RAW data, to which the answer is "no". It's "no"
for the 10D, also; both have full dynamic range at ISO 100.

--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <JPS@no.komm>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
Anonymous
February 14, 2005 12:01:52 AM

Archived from groups: rec.photo.digital.slr-systems (More info?)

On Sun, 13 Feb 2005 21:01:51 GMT, JPS@no.komm wrote:

>I think that drs' question is actually something like, "is the 20D's ISO
>100 actually its ISO 200, pulled one stop", implying a loss of one stop
>of highlights in the RAW data, to which the answer is "no". It's "no"
>for the 10D, also; both have full dynamic range at ISO 100.

Thanks for the info. I know so little about this that I can't word the
question accurately. But you're right, I was referring to a suggestion
that ISO 200 might be less noisy than 100 on the 10D and 20D, and the
person used the term native to describe that best setting.

Another question: Do sensors show a lot of variation between
individual units? i.e. might the one in my 20D be quite different from
another 20D?
Anonymous
February 14, 2005 12:01:53 AM

Archived from groups: rec.photo.digital.slr-systems (More info?)

> Thanks for the info. I know so little about this that I can't word the
> question accurately. But you're right, I was referring to a suggestion
> that ISO 200 might be less noisy than 100 on the 10D and 20D, and the
> person used the term native to describe that best setting.

As a general rule, the lowest ISO will yield the lowest noise and the best
possible signal to noise ratio. The inherent signal to noise ratio of any
sensor is only degraded by adding gain after the sensor. It's a basic
engineering principle. There can be an improvement in signal to noise ratio
by adding gain before a sensor, but never after the sensor. Boosting ISO on
a digital camera simply adds gain after the sensor.

> Another question: Do sensors show a lot of variation between
> individual units? i.e. might the one in my 20D be quite different from
> another 20D?

On average, the difference from sensor to sensor would be minimal. However,
there could be local problems with any given sensor. These include dead
pixels, stuck pixels, etc.
Anonymous
February 14, 2005 4:36:24 AM

Archived from groups: rec.photo.digital.slr-systems (More info?)

In message <86mv01hqphcbqutbj4q7md20mksvpkghg1@4ax.com>,
drs@canby.com wrote:

>Another question: Do sensors show a lot of variation between
>individual units? i.e. might the one in my 20D be quite different from
>another 20D?

I don't know, but if it did, the issue would be obscured by calibration
of the amplifiers. We can't see what the actual levels are in the
sensor, as all we can see is the RAW data that is digitized from it.
Some people's 20Ds seem prone to banding with horizontal bands about 9
pixels aparts (a 9-pixel cycle), but this may have more to do with the
read-out circuitry than the sensor itself.
--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <JPS@no.komm>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
Anonymous
February 14, 2005 2:59:30 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

John McWilliams wrote:

> Alan Browne wrote:
>
>> drs@canby.com wrote:
>>
>>> Is there a way to determine the native ISO of the sensor in my Canon
>>> 20D? From what I've heard, it isn't necessarily the lowest ISO
>>> setting. Someone said the sensor in the 10D was best at 200 and that
>>> 100 was actually a slight interpolation.
>>
>>
>>
>> Native? It's a sensor that 'accumulates' a 'charge' of light over time.
>>
>> For example, it does not care if the exposure time is 1/4000 or 40
>> seconds. At some point there isn't enough light ("zero" accumlation)
>> and at some point there is too much (saturated -> burt highlights).
>>
>> What you might be referring to is "at what ISO setting is the gain set
>> to 1" (0 dB). Even then I doubt the answer is that simple.
>>
>> Best ISO? I believe if you go to dpreview and look at the noise
>> graphs you'll find the D20 is least noisy at ISO 100.
>> http://www.dpreview.com/reviews/konicaminolta7d/page18....
>
>
> Then using the term "native" in this context is quite clear to most. See
> JP Sheehy's reply also.

I did, and JPS' posts are always worth reading attentively.

Unless you have the engineering data and know at what point _all_ the gains are
set to 1, then you don't know what the "native" sensitivity is. And I would bet
strongly that it is not a nicely rounded number like 100 or 200 or 50.

Since the D20 is least noisy at ISO 100 (see link above) that is the sensitivity
one should aim to use, light, lens and time allowing.

In short, the term 'native' just doesn't and can't apply (IMO).

Cheers,
Alan

--
-- r.p.e.35mm user resource: http://www.aliasimages.com/rpe35mmur.htm
-- r.p.d.slr-systems: http://www.aliasimages.com/rpdslrsysur.htm
-- [SI] gallery & rulz: http://www.pbase.com/shootin
-- e-meil: there's no such thing as a FreeLunch.
Anonymous
February 14, 2005 9:10:56 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

Alan Browne <alan.browne@freelunchVideotron.ca> wrote in news:cuqldi$o2v$1
@inews.gazeta.pl:

> In short, the term 'native' just doesn't and can't apply (IMO).

Yes - it can. The native ISO number is determined by the white
clipping level and this is the lowest ISO you can use at
normal contrast pictures without clipping. You can then increase
the ISO by amplifying the signal before A/D conversion until
you no longer can stand the noise.


/Roland
Anonymous
February 14, 2005 9:10:57 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

Roland Karlsson wrote:

> Alan Browne <alan.browne@freelunchVideotron.ca> wrote in news:cuqldi$o2v$1
> @inews.gazeta.pl:
>
>
>>In short, the term 'native' just doesn't and can't apply (IMO).
>
>
> Yes - it can. The native ISO number is determined by the white
> clipping level and this is the lowest ISO you can use at
> normal contrast pictures without clipping. You can then increase
> the ISO by amplifying the signal before A/D conversion until
> you no longer can stand the noise.

The likelyhood that the native sensitivity in ISO is user selectable is
unlikely, and that seems to be what the poster is looking for. As the lowest
ISO yields the lowest noise, that would appear to be the best setting for him
for lowest noise images providing he has the aperture and time to do it.

--
-- r.p.e.35mm user resource: http://www.aliasimages.com/rpe35mmur.htm
-- r.p.d.slr-systems: http://www.aliasimages.com/rpdslrsysur.htm
-- [SI] gallery & rulz: http://www.pbase.com/shootin
-- e-meil: there's no such thing as a FreeLunch.
Anonymous
February 15, 2005 6:58:13 AM

Archived from groups: rec.photo.digital.slr-systems (More info?)

In message <cuqldi$o2v$1@inews.gazeta.pl>,
Alan Browne <alan.browne@freelunchVideotron.ca> wrote:

>Unless you have the engineering data and know at what point _all_ the gains are
>set to 1, then you don't know what the "native" sensitivity is. And I would bet
>strongly that it is not a nicely rounded number like 100 or 200 or 50.

I doubt that it's as straightforward as "there's a voltage is the sensor
itself, and this voltage is used directly at ISO 100, and all other ISOs
have to amplify the signal by 2x, 4x, 8x, etc". They most likely all go
throught the same process, with the 1 of 1x just being a point of
reference, and may actually be 0.13 or 57 or 1311.
--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <JPS@no.komm>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
Anonymous
February 15, 2005 7:24:05 AM

Archived from groups: rec.photo.digital.slr-systems (More info?)

In message <Xns95FDC322A3546klotjohan@130.133.1.4>,
Roland Karlsson <roland_dot_karlsson@bonetmail.com> wrote:

>Alan Browne <alan.browne@freelunchVideotron.ca> wrote in news:cuqldi$o2v$1
>@inews.gazeta.pl:
>
>> In short, the term 'native' just doesn't and can't apply (IMO).
>
>Yes - it can. The native ISO number is determined by the white
>clipping level and this is the lowest ISO you can use at
>normal contrast pictures without clipping.

.... but why not just call this the minimum full-DR ISO? Native sounds
like everything else is artificial, and a host of other connotations,
and so much is this is purely arbitrary. My 20D meters the same as my
Sekonic 558 meter, and I get dark images from it, unless I set the EC to
+1. My 10D meters like the Sekonic at ISO 64, at "ISO 100". Both give
dark images that waste dynamic range in JPEG, and a stop more in RAW.
The 10D can handle it's "ISO 100" shots with a light meter set to ISO 32
with slide-like headroom (with a little extra in the red channel). You
can call any level in the 0-4095 (often only 0-4007 in some ISO modes in
some cameras) "average grey", and render your gamma-corrected image from
there. In fact, that's what all digital cameras do with separate color
channels.

If there were some ISO value that could be ascribed to a white object at
the clipping point of the lowest full-DR ISO, I doubt it would be 100
times a power of 2.

>You can then increase
>the ISO by amplifying the signal before A/D conversion until
>you no longer can stand the noise.

Do you actually think that the sensor data goes straight to the ADC
circuit at "the native ISO", and that the other ISOs go through a
different route with an amplifier that the "native" doesn't pass
through?
--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <JPS@no.komm>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
Anonymous
February 15, 2005 7:36:17 AM

Archived from groups: rec.photo.digital.slr-systems (More info?)

In message <cur4rp$6ji$1@inews.gazeta.pl>,
Alan Browne <alan.browne@freelunchVideotron.ca> wrote:

>The likelyhood that the native sensitivity in ISO is user selectable is
>unlikely, and that seems to be what the poster is looking for. As the lowest
>ISO yields the lowest noise, that would appear to be the best setting for him
>for lowest noise images providing he has the aperture and time to do it.

Of course, many cameras have extensions to the ISO range which are
compromised. Most "ISO 3200" modes on DLRS are just the camera pushing
amplified ISO 1600 to 3200 for you automatically. ISO 50 on the Canon
1D mk II and 1Ds mk II are just ISO 100 pulled for you by the camera,
with a loss of a stop of headroom (I wonder if they posterize the data,
or contain a tag in the RAW file to expose what would otherwise be ISO
100 data a stop darker). I shoot like this on my 10D and 20D all the
time, even though the LCD doesn't say, "ISO 50". It says "ISO 100" and
"+1 EC" instead. On the other end of the spectrum, I don't set my 20D
to ISO 3200 anymore. I set it to ISO 1600 and -1 EC instead, which
gives the same quality in the mutual dynamic range, plus an extra stop
for specular highlights and light sources. The only exception is that
bad pixels are interpolated with higher resolution at "ISO 3200", which
is trivial (all 20D ISO 3200 RAW values are even, except some of the
interpolated bad pixels).

If I am drifting a bit off topic, I just woke up from an evening nap and
don't have my bearings straight yet.
--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <JPS@no.komm>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
Anonymous
February 15, 2005 2:24:03 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

JPS@no.komm wrote:

> In message <cuqldi$o2v$1@inews.gazeta.pl>,
> Alan Browne <alan.browne@freelunchVideotron.ca> wrote:
>
>
>>Unless you have the engineering data and know at what point _all_ the gains are
>>set to 1, then you don't know what the "native" sensitivity is. And I would bet
>>strongly that it is not a nicely rounded number like 100 or 200 or 50.
>
>
> I doubt that it's as straightforward as "there's a voltage is the sensor
> itself, and this voltage is used directly at ISO 100, and all other ISOs
> have to amplify the signal by 2x, 4x, 8x, etc". They most likely all go
> throught the same process, with the 1 of 1x just being a point of
> reference, and may actually be 0.13 or 57 or 1311.

That's really what I was getting to. There may be a point within ISO 100 to
3200 where the gain is 1 (pre A/D), it may be that the unity gain is at ISO
39.07... who knows, and it ain't important in the end.

Cheers,
Alan.


--
-- r.p.e.35mm user resource: http://www.aliasimages.com/rpe35mmur.htm
-- r.p.d.slr-systems: http://www.aliasimages.com/rpdslrsysur.htm
-- [SI] gallery & rulz: http://www.pbase.com/shootin
-- e-meil: there's no such thing as a FreeLunch.
Anonymous
February 15, 2005 2:35:49 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

JPS@no.komm wrote:

> In message <cur4rp$6ji$1@inews.gazeta.pl>,
> Alan Browne <alan.browne@freelunchVideotron.ca> wrote:
>
>
>>The likelyhood that the native sensitivity in ISO is user selectable is
>>unlikely, and that seems to be what the poster is looking for. As the lowest
>>ISO yields the lowest noise, that would appear to be the best setting for him
>>for lowest noise images providing he has the aperture and time to do it.
>
>
> Of course, many cameras have extensions to the ISO range which are
> compromised. Most "ISO 3200" modes on DLRS are just the camera pushing
> amplified ISO 1600 to 3200 for you automatically. ISO 50 on the Canon
> 1D mk II and 1Ds mk II are just ISO 100 pulled for you by the camera,
> with a loss of a stop of headroom (I wonder if they posterize the data,
> or contain a tag in the RAW file to expose what would otherwise be ISO
> 100 data a stop darker). I shoot like this on my 10D and 20D all the
> time, even though the LCD doesn't say, "ISO 50". It says "ISO 100" and
> "+1 EC" instead. On the other end of the spectrum, I don't set my 20D

in "100 +1 EC" is the shutter period increased ("A" mode) or is the image sensor
'gained'? If it just increases the exposure time, then it has nothing to do
with the sensor gain.

> to ISO 3200 anymore. I set it to ISO 1600 and -1 EC instead, which
> gives the same quality in the mutual dynamic range, plus an extra stop
> for specular highlights and light sources. The only exception is that
> bad pixels are interpolated with higher resolution at "ISO 3200", which
> is trivial (all 20D ISO 3200 RAW values are even, except some of the
> interpolated bad pixels).

This suggests that you're simply over/underexposing, not actually changing the
gain of the sensor. Eg, you're managing the limit cases optimally knowing what
you're likely to get during RAW conversion.

>
> If I am drifting a bit off topic, I just woke up from an evening nap and
> don't have my bearings straight yet.

Oil your bearings. You're drifting a bit, but hey, it's usenet.

Cheers,
Alan
--
-- r.p.e.35mm user resource: http://www.aliasimages.com/rpe35mmur.htm
-- r.p.d.slr-systems: http://www.aliasimages.com/rpdslrsysur.htm
-- [SI] gallery & rulz: http://www.pbase.com/shootin
-- e-meil: there's no such thing as a FreeLunch.
Anonymous
February 15, 2005 6:15:12 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

JPS@no.komm wrote:
> In message <cuqldi$o2v$1@inews.gazeta.pl>,
> Alan Browne <alan.browne@freelunchVideotron.ca> wrote:
>
>
>>Unless you have the engineering data and know at what point _all_ the gains are
>>set to 1, then you don't know what the "native" sensitivity is. And I would bet
>>strongly that it is not a nicely rounded number like 100 or 200 or 50.
>
>
> I doubt that it's as straightforward as "there's a voltage is the sensor
> itself, and this voltage is used directly at ISO 100, and all other ISOs
> have to amplify the signal by 2x, 4x, 8x, etc". They most likely all go
> throught the same process, with the 1 of 1x just being a point of
> reference, and may actually be 0.13 or 57 or 1311.

One thing that is being overlooked in the ISO rating is that 20Ds (just
the same as 10Ds before them) lose definition at higher ISO settings
until it is impossible to see detail in an otherwise sharply focused
image at 800 and higher ISO when you look at the actual pixel size on a
monitor. Sometimes the image will print 'looking' sharper but not
anywhere near as sharp as a 100 ISO similar shot. Just another reason to
use fast glass with digitals.

JD

--
EOS my GOD,
Give me ISO for I have not yet seen the light.
Take away my grain, give me colour and you
shall have given me the edge!
Anonymous
February 15, 2005 6:15:13 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

In message <42118600$1@dnews.tpgi.com.au>,
Deciple of EOS <decipleofeos@yahoo.com> wrote:

>One thing that is being overlooked in the ISO rating is that 20Ds (just
>the same as 10Ds before them) lose definition at higher ISO settings
>until it is impossible to see detail in an otherwise sharply focused
>image at 800 and higher ISO when you look at the actual pixel size on a
>monitor.

Who is overlooking that? It's not mentioned often, because it is taken
for granted; all other things being equal, higher ISOs are noisier and
therefore harder to see pixel-to-pixel detail in.

>Sometimes the image will print 'looking' sharper but not
>anywhere near as sharp as a 100 ISO similar shot. Just another reason to
>use fast glass with digitals.

True, but not all fast lenses are sharp wide open, nor is the depth of
field sufficient for all uses.

People who avoid high ISOs at all costs often ruin their pictures. In
DPReview, just last night, someone started a thread asking why his ISO
100 shot of a gull on a beach was so "noisy". I looked at the image,
and it looked more like over-quantization than sensor noise. I then
looked at the EXIF data, and it was shot at 1/8000 @ f/2.8. That's
sunny-f/25, and I know that you need sunny-f/11 for a decent exposure on
the 10D or 20D (and it was obviously taken uner a hazy sky, so sunny-f/8
would probably be better). The image was under-exposed by about 5
stops, probably using no more than 4 to 6 of the 12 bits in the RAW
data. Maybe he thought that ISO 100 would give him the best picture,
just a bit too hard.
--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <JPS@no.komm>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
Anonymous
February 15, 2005 6:15:14 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

On Tue, 15 Feb 2005 14:59:18 GMT, JPS@no.komm wrote:

>People who avoid high ISOs at all costs often ruin their pictures. In
>DPReview, just last night, someone started a thread asking why his ISO
>100 shot of a gull on a beach was so "noisy". I looked at the image,
>and it looked more like over-quantization than sensor noise. I then
>looked at the EXIF data, and it was shot at 1/8000 @ f/2.8. That's
>sunny-f/25, and I know that you need sunny-f/11 for a decent exposure on
>the 10D or 20D (and it was obviously taken uner a hazy sky, so sunny-f/8
>would probably be better). The image was under-exposed by about 5
>stops, probably using no more than 4 to 6 of the 12 bits in the RAW
>data. Maybe he thought that ISO 100 would give him the best picture,
>just a bit too hard.

After reading all these posts, I realize that the notion of a "native"
ISO is at least far more complicated than I thought. And possibly a
misnomer. The idea that ISO 200 on the 20D might produce less noise
than any other setting is incorrect. So I'm back to keeping it simple:
for reduced noise, shoot at as low an ISO as possible, which is what I
thought before I heard about native ISO. But now I have to look up
"over-quantization."
Anonymous
February 15, 2005 7:12:45 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

JPS@no.komm wrote:
> Over-quantization refers to the state of digital data after a case of
> under-exposure. The DSLRs are capable of recording about 3877 to
3965
> distinct levels of subject brightness for each pixel. The cameras
> generally expose things so that midtones are in the hundreds (300 to
600
> or so), and highlights up to about 2000.

snip

> <>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
> John P Sheehy <JPS@no.komm>
> ><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><

John,

Canon digitals are generating 12 bits (4096 levels) out of the analog
to digital converter, but that's for *each* of the red, green, and blue
sensor 'wells'. I'm unsure exactly how the RAW decoding works, but
suspect that brightness levels of the *decoded* image pixel is far
larger than 12 bits.

Rodger Clark (clarkvision.com) has done some great research on this
subject.
Anonymous
February 15, 2005 8:05:44 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

In message <cut8d3$lfr$1@inews.gazeta.pl>,
Alan Browne <alan.browne@freelunchVideotron.ca> wrote:

>JPS@no.komm wrote:
>
>> In message <cur4rp$6ji$1@inews.gazeta.pl>,
>> Alan Browne <alan.browne@freelunchVideotron.ca> wrote:
>>
>>
>>>The likelyhood that the native sensitivity in ISO is user selectable is
>>>unlikely, and that seems to be what the poster is looking for. As the lowest
>>>ISO yields the lowest noise, that would appear to be the best setting for him
>>>for lowest noise images providing he has the aperture and time to do it.
>>
>>
>> Of course, many cameras have extensions to the ISO range which are
>> compromised. Most "ISO 3200" modes on DLRS are just the camera pushing
>> amplified ISO 1600 to 3200 for you automatically. ISO 50 on the Canon
>> 1D mk II and 1Ds mk II are just ISO 100 pulled for you by the camera,
>> with a loss of a stop of headroom (I wonder if they posterize the data,
>> or contain a tag in the RAW file to expose what would otherwise be ISO
>> 100 data a stop darker). I shoot like this on my 10D and 20D all the
>> time, even though the LCD doesn't say, "ISO 50". It says "ISO 100" and
>> "+1 EC" instead. On the other end of the spectrum, I don't set my 20D
>
>in "100 +1 EC" is the shutter period increased ("A" mode) or is the image sensor
>'gained'? If it just increases the exposure time, then it has nothing to do
>with the sensor gain.

True, but is that really a useful way to look at it? "100 +1EC" is
actually a better ISO 50 than if the camera did ISO 50 as the "lowest
full-DR ISO", if the lost stop of headroom were not needed. The DR that
you *do* have is quantized 2x finer, and the intensity of the analog
noise in the sensor is the same, and amplfier noise differences at these
ISOs is negligible. The logistics of this are interesting, if you
consider the fact that if you shot JPEG instead of RAW, you would lose
this stop, anyway!

>> to ISO 3200 anymore. I set it to ISO 1600 and -1 EC instead, which
>> gives the same quality in the mutual dynamic range, plus an extra stop
>> for specular highlights and light sources. The only exception is that
>> bad pixels are interpolated with higher resolution at "ISO 3200", which
>> is trivial (all 20D ISO 3200 RAW values are even, except some of the
>> interpolated bad pixels).

>This suggests that you're simply over/underexposing, not actually changing the
>gain of the sensor. Eg, you're managing the limit cases optimally knowing what
>you're likely to get during RAW conversion.

In this particular instance, it is a win-win situation. Pushing 1600 to
3200 doesn't lose a stop of DR; it gains a stop of DR headroom,
actually, over using the camera's "ISO 3200". In pulling 100 to 50,
there actually is a compromise involved; the loss of a stop of DR in the
headroom (which you would never see in JPEG mode with normal contrast
settings, anyway).

--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <JPS@no.komm>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
Anonymous
February 15, 2005 8:05:45 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

JPS@no.komm wrote:

> In message <cut8d3$lfr$1@inews.gazeta.pl>,
> Alan Browne <alan.browne@freelunchVideotron.ca> wrote:
>
>
>>JPS@no.komm wrote:
>>
>>
>>>In message <cur4rp$6ji$1@inews.gazeta.pl>,
>>>Alan Browne <alan.browne@freelunchVideotron.ca> wrote:
>>>
>>>
>>>
>>>>The likelyhood that the native sensitivity in ISO is user selectable is
>>>>unlikely, and that seems to be what the poster is looking for. As the lowest
>>>>ISO yields the lowest noise, that would appear to be the best setting for him
>>>>for lowest noise images providing he has the aperture and time to do it.
>>>
>>>
>>>Of course, many cameras have extensions to the ISO range which are
>>>compromised. Most "ISO 3200" modes on DLRS are just the camera pushing
>>>amplified ISO 1600 to 3200 for you automatically. ISO 50 on the Canon
>>>1D mk II and 1Ds mk II are just ISO 100 pulled for you by the camera,
>>>with a loss of a stop of headroom (I wonder if they posterize the data,
>>>or contain a tag in the RAW file to expose what would otherwise be ISO
>>>100 data a stop darker). I shoot like this on my 10D and 20D all the
>>>time, even though the LCD doesn't say, "ISO 50". It says "ISO 100" and
>>>"+1 EC" instead. On the other end of the spectrum, I don't set my 20D
>>
>>in "100 +1 EC" is the shutter period increased ("A" mode) or is the image sensor
>>'gained'? If it just increases the exposure time, then it has nothing to do
>>with the sensor gain.
>
>
> True, but is that really a useful way to look at it? "100 +1EC" is

That "was" the subject. I'm not criiticizing what you're doing/suggesting as
you've explored the extremes of the sensors more than most people and you know
where the quality limits are. You should write a webpage on it.

> actually a better ISO 50 than if the camera did ISO 50 as the "lowest
> full-DR ISO", if the lost stop of headroom were not needed. The DR that
> you *do* have is quantized 2x finer, and the intensity of the analog
> noise in the sensor is the same, and amplfier noise differences at these
> ISOs is negligible. The logistics of this are interesting, if you
> consider the fact that if you shot JPEG instead of RAW, you would lose
> this stop, anyway!

Sounds good.

>
>
>>>to ISO 3200 anymore. I set it to ISO 1600 and -1 EC instead, which
>>>gives the same quality in the mutual dynamic range, plus an extra stop
>>>for specular highlights and light sources. The only exception is that
>>>bad pixels are interpolated with higher resolution at "ISO 3200", which
>>>is trivial (all 20D ISO 3200 RAW values are even, except some of the
>>>interpolated bad pixels).
>
>
>>This suggests that you're simply over/underexposing, not actually changing the
>>gain of the sensor. Eg, you're managing the limit cases optimally knowing what
>>you're likely to get during RAW conversion.
>
>
> In this particular instance, it is a win-win situation. Pushing 1600 to
> 3200 doesn't lose a stop of DR; it gains a stop of DR headroom,
> actually, over using the camera's "ISO 3200". In pulling 100 to 50,
> there actually is a compromise involved; the loss of a stop of DR in the
> headroom (which you would never see in JPEG mode with normal contrast
> settings, anyway).

As above.

Cheers,
Alan


--
-- r.p.e.35mm user resource: http://www.aliasimages.com/rpe35mmur.htm
-- r.p.d.slr-systems: http://www.aliasimages.com/rpdslrsysur.htm
-- [SI] gallery & rulz: http://www.pbase.com/shootin
-- e-meil: there's no such thing as a FreeLunch.
Anonymous
February 15, 2005 8:31:46 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

In message <cut7n1$i6m$1@inews.gazeta.pl>,
Alan Browne <alan.browne@freelunchVideotron.ca> wrote:

>That's really what I was getting to. There may be a point within ISO 100 to
>3200 where the gain is 1 (pre A/D), it may be that the unity gain is at ISO
>39.07... who knows, and it ain't important in the end.

People need to drop the notion, though, that the lowest ISO in the
camera is the real one, and that the others are of a totally different
quality. You don't know how many people are going around with their
cameras permanently fixed to ISO 100, taking under-exposed,
overly-shallow-DOF, and motion-blurred pictures because of the "Best and
Native ISO" concept. Since we are stuck with 12-bit quantization on
these cameras, the usefulness of the lowest ISO is rather limited. The
quality difference between ISO 100 and 400 would be *much* greater if
the digitization were 16-bit instead of 12. 12-bit cripples ISO 100, in
terms of useable shadows and under-exposure. If it were 16-bit, a shot
under-exposed by 4 stops at ISO 100 would look as good as ISO 1600 does
now, with 12-bit capture.

I see the capture on the sensor of my DSLRs as totally dependent on
lighting, aperture, and shutter speed. ISO has no effect on this at
all, except in how it affects metering, and therefore, exposure.

I see the ISO setting on my DSLRs as a relative amount of amplification
of the signal, and therefore a scaler of the 12-bit data range that the
analog sensor information is digitized to. The very highest
amplifications add visible noise and distortion of their own.
--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <JPS@no.komm>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
Anonymous
February 15, 2005 8:31:47 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

JPS@no.komm wrote:

> In message <cut7n1$i6m$1@inews.gazeta.pl>,
> Alan Browne <alan.browne@freelunchVideotron.ca> wrote:
>
>
>>That's really what I was getting to. There may be a point within ISO 100 to
>>3200 where the gain is 1 (pre A/D), it may be that the unity gain is at ISO
>>39.07... who knows, and it ain't important in the end.
>
>
> People need to drop the notion, though, that the lowest ISO in the
> camera is the real one, and that the others are of a totally different
> quality. You don't know how many people are going around with their
> cameras permanently fixed to ISO 100, taking under-exposed,
> overly-shallow-DOF, and motion-blurred pictures because of the "Best and
> Native ISO" concept. Since we are stuck with 12-bit quantization on

Hmm. And you objected to my "others" statment? <g>. Photography remains
photography. As I said earlier in this thread:

"Since the D20 is least noisy at ISO 100 (see link above) that is the
sensitivity one should aim to use, light, lens and time allowing."

I shoud add, ", if of course low noise images are essential."

All of my lenses are /2.8 or faster. Most DSLR owners have slower lenses, so
higher ISO needs to be considered as acceptable.

> these cameras, the usefulness of the lowest ISO is rather limited. The
> quality difference between ISO 100 and 400 would be *much* greater if
> the digitization were 16-bit instead of 12. 12-bit cripples ISO 100, in
> terms of useable shadows and under-exposure. If it were 16-bit, a shot
> under-exposed by 4 stops at ISO 100 would look as good as ISO 1600 does
> now, with 12-bit capture.

Higher cost A/D, but shouldn't be an obstacle (they did it in scanners). Maybe
14 bit as a compromise. I believer higher bit A/D also consumes more juice,
another trade-off.

Extending the dynamic range 'down' into 14 and 16 bits would reveal more noise.
Perhaps to a point where the detail 'gained' looks so bad as to be better off
hidden.

>
> I see the capture on the sensor of my DSLRs as totally dependent on
> lighting, aperture, and shutter speed. ISO has no effect on this at
> all, except in how it affects metering, and therefore, exposure.

Yep.
>
> I see the ISO setting on my DSLRs as a relative amount of amplification
> of the signal, and therefore a scaler of the 12-bit data range that the
> analog sensor information is digitized to. The very highest
> amplifications add visible noise and distortion of their own.

Yep. Quantization noise, itself made even worse by the noise in the image
having quantization noise. (noise-noise?). At least the pixels are constrained
in x,y unlike film.

Cheers,
Alan.


--
-- r.p.e.35mm user resource: http://www.aliasimages.com/rpe35mmur.htm
-- r.p.d.slr-systems: http://www.aliasimages.com/rpdslrsysur.htm
-- [SI] gallery & rulz: http://www.pbase.com/shootin
-- e-meil: there's no such thing as a FreeLunch.
Anonymous
February 16, 2005 12:58:30 AM

Archived from groups: rec.photo.digital.slr-systems (More info?)

JPS@no.komm wrote in news:3vs211pe421vikbfjst17jdqnukpd7bgrd@4ax.com:

> You can then increase
>>the ISO by amplifying the signal before A/D conversion until
>>you no longer can stand the noise.
>
> Do you actually think that the sensor data goes straight to the ADC
> circuit at "the native ISO", and that the other ISOs go through a
> different route with an amplifier that the "native" doesn't pass
> through?

Nope.

I am talking about the sensor ISO - and that has nothing to do
with A/D conversion. That is a strictly analog definition. It
is based upon some stops below saturation of the sensor.

When you do the A/D conversion it is not really any idea to
go below that ISO.


/Roland
Anonymous
February 16, 2005 2:38:06 AM

Archived from groups: rec.photo.digital.slr-systems (More info?)

In message <7p8411t8q4om4hi5f773f483rnqv2jbhjr@4ax.com>,
drs@canby.com wrote:

>On Tue, 15 Feb 2005 14:59:18 GMT, JPS@no.komm wrote:
>
>>People who avoid high ISOs at all costs often ruin their pictures. In
>>DPReview, just last night, someone started a thread asking why his ISO
>>100 shot of a gull on a beach was so "noisy". I looked at the image,
>>and it looked more like over-quantization than sensor noise. I then
>>looked at the EXIF data, and it was shot at 1/8000 @ f/2.8. That's
>>sunny-f/25, and I know that you need sunny-f/11 for a decent exposure on
>>the 10D or 20D (and it was obviously taken uner a hazy sky, so sunny-f/8
>>would probably be better). The image was under-exposed by about 5
>>stops, probably using no more than 4 to 6 of the 12 bits in the RAW
>>data. Maybe he thought that ISO 100 would give him the best picture,
>>just a bit too hard.
>
>After reading all these posts, I realize that the notion of a "native"
>ISO is at least far more complicated than I thought. And possibly a
>misnomer. The idea that ISO 200 on the 20D might produce less noise
>than any other setting is incorrect. So I'm back to keeping it simple:
>for reduced noise, shoot at as low an ISO as possible, which is what I
>thought before I heard about native ISO. But now I have to look up
>"over-quantization."

Over-quantization refers to the state of digital data after a case of
under-exposure. The DSLRs are capable of recording about 3877 to 3965
distinct levels of subject brightness for each pixel. The cameras
generally expose things so that midtones are in the hundreds (300 to 600
or so), and highlights up to about 2000. If you under-expose by 5
stops, you divide these values by 32 to get 9 to 18 for the midtones,
and highlights at about 60. That is only about 6 bits of information
for each pixel. It will appear dark and under-exposed, until you boost
the levels in the RAW converter and/or the image editor, and instead of
smooth gradations, you will have cartoon-like color, broken up only by
noise.
--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <JPS@no.komm>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
Anonymous
February 17, 2005 12:39:06 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

skid00skid00@yahoo.com wrote:
> JPS@no.komm wrote:
> > Over-quantization refers to the state of digital data after a case
of
> > under-exposure. The DSLRs are capable of recording about 3877 to
> 3965
> > distinct levels of subject brightness for each pixel. The cameras
> > generally expose things so that midtones are in the hundreds (300
to
> 600
> > or so), and highlights up to about 2000.
>
> Canon digitals are generating 12 bits (4096 levels) out of the analog
> to digital converter

His practical 3965 is very close to your theoretical 4096.

> but that's for *each* of the red, green, and blue
> sensor 'wells'. I'm unsure exactly how the RAW decoding works, but
> suspect that brightness levels of the *decoded* image pixel is far
> larger than 12 bits.

The capture has 12 bits of linear data, no matter how you decode or
process it. After gamma, some range will need more than 12 bits to keep
the original 12 bits, but other ranges will have far less than 12 bits
resolution. The overall distinct levels are still limited by 4096.
Anonymous
February 19, 2005 6:55:42 PM

Archived from groups: rec.photo.digital.slr-systems (More info?)

On Tue, 15 Feb 2005 14:59:18 GMT, JPS@no.komm <JPS@no.komm> wrote:
>
> People who avoid high ISOs at all costs often ruin their pictures. In
> DPReview, just last night, someone started a thread asking why his ISO
> 100 shot of a gull on a beach was so "noisy". I looked at the image,
> and it looked more like over-quantization than sensor noise. I then
> looked at the EXIF data, and it was shot at 1/8000 @ f/2.8.

You think ISO is the problem here? Or shutter speed?

--
Ben Rosengart (212) 741-4400 x215
Sometimes it only makes sense to focus our attention on those
questions that are equal parts trivial and intriguing.
--Josh Micah Marshall
Anonymous
February 20, 2005 1:51:19 AM

Archived from groups: rec.photo.digital.slr-systems (More info?)

In message <slrnd1eobu.s6n.br@panix5.panix.com>,
Ben Rosengart <br+rpdss@panix.com> wrote:

>On Tue, 15 Feb 2005 14:59:18 GMT, JPS@no.komm <JPS@no.komm> wrote:
>>
>> People who avoid high ISOs at all costs often ruin their pictures. In
>> DPReview, just last night, someone started a thread asking why his ISO
>> 100 shot of a gull on a beach was so "noisy". I looked at the image,
>> and it looked more like over-quantization than sensor noise. I then
>> looked at the EXIF data, and it was shot at 1/8000 @ f/2.8.

>You think ISO is the problem here? Or shutter speed?

It seemed at the time that the shutter speed was intentional. He later
said that it was the result of cneter-weighted metering on the gull's
white breast, but I find that bizarre, still. In such a context, a
higher ISO would have resulted in much less posterization with the same
aperture and shutter speed.

He could have done this shot right at ISO 100.
--

<>>< ><<> ><<> <>>< ><<> <>>< <>>< ><<>
John P Sheehy <JPS@no.komm>
><<> <>>< <>>< ><<> <>>< ><<> ><<> <>><
!