Can I see those high ISO pictures again?

I thought I had archived a certain thread about noise in high ISO
pictures. Someone posted a very good "article" which contained two enlarged
pictures of a piling shot at a low and very high ISO. My memory is hazy
about the post but it seemed to show that the ISO 100 photo was noisier than
the high ISO photo. But there was a special way it had been taken.

Does anyone remember that post? Can anyone point me to the thread? It
was posted within the last few months. I've already started going through
Google, but, it's slow going.

TIA


Steve

Steve wrote:

I thought I had archived a certain thread about noise in high ISO
pictures. Someone posted a very good "article" which contained two enlarged
pictures of a piling shot at a low and very high ISO. My memory is hazy
about the post but it seemed to show that the ISO 100 photo was noisier than
the high ISO photo. But there was a special way it had been taken.

Does anyone remember that post? Can anyone point me to the thread? It
was posted within the last few months. I've already started going through
Google, but, it's slow going.

That was by JPS. The idea is to overexpose the high ISO version (as long
as there aren't highlights to blow) then darken it in photoshop. Simply
maxing the exposure is what improves the quality. If the light is bright
enough, you don't need to boost ISO, just EC (Exposure Compensation).

Here's my experiment on that question using a D70:
http://www.edgehill.net/1/?SC=go.php...y/expose-right

Google the term 'expose right' I think www.luminouslandscape.com has a
good article explaining the math.

In message ,
"Steve" wrote:

I thought I had archived a certain thread about noise in high ISO
pictures. Someone posted a very good "article" which contained two enlarged
pictures of a piling shot at a low and very high ISO. My memory is hazy
about the post but it seemed to show that the ISO 100 photo was noisier than
the high ISO photo. But there was a special way it had been taken.

Does anyone remember that post? Can anyone point me to the thread? It
was posted within the last few months. I've already started going through
Google, but, it's slow going.

That was me:

http://www.pbase.com/jps_photo/image/57149515

The purpose of the shot was to show how bad the deep shadows are at ISO
100, and how the camera, the Canon 20D, and most recent Canons are
actually optimized for ISO 1600, as it has the cleanest amplification
(but of course, it only amplifies the lowest 1/16 of the sensor's
dynamic range).

Both shots (100% crops, actually) were taken with the same f-stop and
shutter speed, in manual exposure mode, and the only difference was the
ISO setting on the camera. One was 100, and the other, 1600. The ISO
1600 image was taken first, and the area of the crop was under-exposed
by 2.6 stops, or an exposure index (push to) ISO 10,000. The ISO 100
was under-exposed by 6.6 stops, and of course, with the same f-stop and
shutter speed, the exposure index was also 10,000.

It seems immediately obvious that under-exposing an image is bad, but
often that thinking comes from experience with film or RAW converters
that don't bring up the shadows very well. With digital, we should
ideally only expect two problems with under-exposure; amplification of
unavoidable sensor noise, and quantization (too few levels to represent
the tones). Both images received the same level of exposure on the
sensor; the only difference is the way that the signals were amplified
off the sensor and turned into numbers, so noise from the photosites
themselves (photon counting noise and dark current noise) should be the
same. The ISO 100 crop is rendered with only the 20 lowest RAW levels
(that's what they used, out of almost 4000 available), and the ISO 1600
image used 320 levels (16x as many), because the ISO 1600 image was
amplified 16x as much as the 100. I didn't put it on the web page, and
perhaps I should add it, but I also intentionally reduced the 1600 crop
to 20 levels on my computer, and rendered it the same way as the 100
crop, to equalize them in terms of quantization, and they were still
very different; the posterization only made the 1600 crop slightly
worse; I had to look for the differences. So that leaves us with one
remaining psossibility - the ISO 100 RAW files created by the camera
have extremely poor shadows, full of noise created in the readout
process. The fact is, even though ISO 1600 amplifies the sensor signals
16x as much as ISO 100, the average noise is only 2.27x as high as ISO
100. There is even less of a difference between 100 and 400; 400 has 4x
the amplification of 100, but only 1.14x as much noise.

The implications are that even with 12-bit RAW data, the ISO 100 images
could have a couple stops more dynamic range than they currently do with
cleaner readout/amplification technology. Roger Clarke, another poster
here, has web pages dedicated to experiments where he shows that ISO
1600 on his Canon 1DmkII is mostly inevitable sensor noise, and very
little readout noise. My estimated guess is that at ISO 100, over 90%
of the shadow noise (other than that caused by quantization) is created
in the readout process, and at ISO 1600, only about 20% of the noise is
created in the readout process.

This makes me wonder if the manufacturer isn't keeping higher DR at ISO
100 up its sleeve to make us buy more cameras later.

The 5D is Canon's latest hi-end imager, and it still has weak shadows at
ISO 100 also, although slightly better than the 20D. The best ISO 100
shadows I've seen were from the Sigma DSLRs (notably in B&W mode), but
they have other issues that make me lose interest.
--


John P Sheehy

wrote in message
...
In message ,
"Steve" wrote:

I thought I had archived a certain thread about noise in high ISO
pictures. Someone posted a very good "article" which contained two
enlarged
pictures of a piling shot at a low and very high ISO. My memory is hazy
about the post but it seemed to show that the ISO 100 photo was noisier
than
the high ISO photo. But there was a special way it had been taken.

Does anyone remember that post? Can anyone point me to the thread?
It
was posted within the last few months. I've already started going through
Google, but, it's slow going.

That was me:

http://www.pbase.com/jps_photo/image/57149515

The purpose of the shot was to show how bad the deep shadows are at ISO
100, and how the camera, the Canon 20D, and most recent Canons are
actually optimized for ISO 1600, as it has the cleanest amplification
(but of course, it only amplifies the lowest 1/16 of the sensor's
dynamic range).

Is there a simple way to determine a camera's dynamic range? Does ISO
1600 relate directly to1/16 of the DR, or is it chance they both have 16 in
them? I mean if my camera is set at ISO 800, am I amplifying the lowest 1/8
of the sensor's DR?


Both shots (100% crops, actually) were taken with the same f-stop and
shutter speed, in manual exposure mode, and the only difference was the
ISO setting on the camera. One was 100, and the other, 1600. The ISO
1600 image was taken first, and the area of the crop was under-exposed
by 2.6 stops, or an exposure index (push to) ISO 10,000. The ISO 100
was under-exposed by 6.6 stops, and of course, with the same f-stop and
shutter speed, the exposure index was also 10,000.

How did you under-expose the ISO 100 photo by 6.6 stops? If you had the same
f-stop and shutter speed, what did you change on the camera? My camera only
does 3 stops of EC, does the 20D do more?


It seems immediately obvious that under-exposing an image is bad, but
often that thinking comes from experience with film or RAW converters
that don't bring up the shadows very well. With digital, we should
ideally only expect two problems with under-exposure; amplification of
unavoidable sensor noise, and quantization (too few levels to represent
the tones). Both images received the same level of exposure on the
sensor; the only difference is the way that the signals were amplified
off the sensor and turned into numbers, so noise from the photosites
themselves (photon counting noise and dark current noise) should be the
same. The ISO 100 crop is rendered with only the 20 lowest RAW levels
(that's what they used, out of almost 4000 available), and the ISO 1600
image used 320 levels (16x as many), because the ISO 1600 image was
amplified 16x as much as the 100. I didn't put it on the web page, and

Are these "RAW levels" in a RAW converter?

perhaps I should add it, but I also intentionally reduced the 1600 crop
to 20 levels on my computer, and rendered it the same way as the 100
crop, to equalize them in terms of quantization, and they were still
very different; the posterization only made the 1600 crop slightly
worse; I had to look for the differences. So that leaves us with one
remaining psossibility - the ISO 100 RAW files created by the camera
have extremely poor shadows, full of noise created in the readout
process. The fact is, even though ISO 1600 amplifies the sensor signals
16x as much as ISO 100, the average noise is only 2.27x as high as ISO
100. There is even less of a difference between 100 and 400; 400 has 4x
the amplification of 100, but only 1.14x as much noise.

Are these noise "multipliers" drawn from the Std Dev numbers for ISO's in
your original post?

ISO Std Dev

100 2.1
200 2.2
400 2.4
800 3.2
1600 4.7
Can I get them for my camera to see if the same holds true?

That's all of my questions... for nowg This is interesting stuff and I
appreciate your very comprehensive posts.


The implications are that even with 12-bit RAW data, the ISO 100 images
could have a couple stops more dynamic range than they currently do with
cleaner readout/amplification technology. Roger Clarke, another poster
here, has web pages dedicated to experiments where he shows that ISO
1600 on his Canon 1DmkII is mostly inevitable sensor noise, and very
little readout noise. My estimated guess is that at ISO 100, over 90%
of the shadow noise (other than that caused by quantization) is created
in the readout process, and at ISO 1600, only about 20% of the noise is
created in the readout process.

This makes me wonder if the manufacturer isn't keeping higher DR at ISO
100 up its sleeve to make us buy more cameras later.

The 5D is Canon's latest hi-end imager, and it still has weak shadows at
ISO 100 also, although slightly better than the 20D. The best ISO 100
shadows I've seen were from the Sigma DSLRs (notably in B&W mode), but
they have other issues that make me lose interest.
--


John P Sheehy

Steve wrote:

Is there a simple way to determine a camera's dynamic range? Does ISO
1600 relate directly to1/16 of the DR, or is it chance they both have 16 in
them? I mean if my camera is set at ISO 800, am I amplifying the lowest 1/8
of the sensor's DR?

Procedures for Evaluating Digital Camera
Sensor Noise, Dynamic Range, and Full Well Capacities;
Canon 1D Mark II Analysis

http://www.clarkvision.com/imagedetail/evaluation-1d2

Roger