"Exposure" vs "Digitization

In message ,
"David J Taylor"

wrote:

Mike Warren wrote:
David J Taylor wrote:
Perhaps there is a simple "double the output of the converter"
happening in cameras where this quantisation is observed? Perhaps
both methods are used to get two extra ISO steps?

Yuk, I hope not.

It would be possible to test this but I'd rather take photos :-)

-Mike

It may be that the realistic maximum speed is ISO 800, but marketing
demands ISO 1600 and ISO 3200. Nothing extra to be extracted from the
sensor, so just double the ADC output and perhaps dither it a little to
disguise the action?

That's what the Canon 10D does, although it seems that its 3200 is 1600
pushed, and its 1600 is 800 pushed, so 3200 is not totally redundant, as
it has the only "1600-level" amplification. The dithering is enough to
fool a histogram, but if you look at the pattern of the least
significant bit, it is obvious:

http://www.pbase.com/jps_photo/image/38841732/original

--


John P Sheehy

wrote:
In message ,
"David J Taylor"

wrote:

Mike Warren wrote:
David J Taylor wrote:
Perhaps there is a simple "double the output of the converter"
happening in cameras where this quantisation is observed? Perhaps
both methods are used to get two extra ISO steps?

Yuk, I hope not.

It would be possible to test this but I'd rather take photos :-)

-Mike

It may be that the realistic maximum speed is ISO 800, but marketing
demands ISO 1600 and ISO 3200. Nothing extra to be extracted from
the sensor, so just double the ADC output and perhaps dither it a
little to disguise the action?

That's what the Canon 10D does, although it seems that its 3200 is
1600 pushed, and its 1600 is 800 pushed, so 3200 is not totally
redundant, as it has the only "1600-level" amplification. The
dithering is enough to fool a histogram, but if you look at the
pattern of the least significant bit, it is obvious:

http://www.pbase.com/jps_photo/image/38841732/original

If it really were dithering, I'm sure they could do much better than that!
They should use a dither pattern which isn't correlated with spatial
position or image brightness level.

Thanks, John.

David

wrote:
In message ,
Alan Browne wrote:

A good point, however a closeup examination of photographs at 1600
and 3200 show discrete pixel jumps that are similar in nature to the
effects of quantization noise. It is (as you say) disguised by
other things in the chain (ADC, unknown local pre-ADC amps), but it
does indeed bear resemblance to quantization noise.

If you want to see noise that is mainly quantization from the
digitization, just crank up the shadows of an under-exposed ISO 100
image. The high-ISO shadows are a combination of that and sensor
noise with a tad of amplifier noise.

Yes, of course, the darkest values at ISO 100 would have the least photon
noise and therefore be the best places to see quantisation noise.

It still surprises me a little: back of the envelope sum: well capacity
32K, quantised to 4K values, so each quantisation step is 8 levels. Scene
dynamic range 1000:1 so smallest value about 32. Photo-limited noise sqrt
(32) is 5-6 levels. OK, so the quantisation step under those extreme
conditions is of the same order as the photon-limited noise. No longer
surprised!

David

wrote in message
...
In message ,
Gregory Blank wrote:

In article ,
wrote:

To answer you quite directly: for lack of using a better description
and to be concise.

If someone decides that "ISO 100 gives the best quality" and gets an
image that utilizes only 1/16th of the RAW values available, they would
have had a much better image if they had the camera set to ISO 1600
with
the same aperture and shutter speed. I have a hard time saying that
they "under-exposed" the image; it makes more sense to say that they
under-digitized it (quantized it) by using too low of an ISO.


Is it just me, or are you getting the ISO's wrong here? ISO 100 allows for
the greatest amount of exposure and equals the best possible image quality
(except for on film where you can go much lower). People aren't being
mislead, it's exactly the same as in the film world. Faster films/iso's
means better low light/hand held photographs, but at the expense of
grain/noise.

--
Regards,
Matt Clara
www.mattclara.com

"Matt Clara" wrote in message
.. .
wrote in message
...
In message ,
Gregory Blank wrote:

In article ,
wrote:

To answer you quite directly: for lack of using a better description
and to be concise.

If someone decides that "ISO 100 gives the best quality" and gets an
image that utilizes only 1/16th of the RAW values available, they
would
have had a much better image if they had the camera set to ISO 1600
with
the same aperture and shutter speed. I have a hard time saying that
they "under-exposed" the image; it makes more sense to say that they
under-digitized it (quantized it) by using too low of an ISO.


Is it just me, or are you getting the ISO's wrong here? ISO 100 allows
for
the greatest amount of exposure and equals the best possible image quality
(except for on film where you can go much lower). People aren't being
mislead, it's exactly the same as in the film world. Faster films/iso's
means better low light/hand held photographs, but at the expense of
grain/noise.


JPS has discussed this many times in r.p.d.s-s. If you Google a little in
r.p.d.s-s. for noise, ISO, and his name as author, you'll see many of his
posts discussing ISO 100 and noise.

Greg

In message ,
"Matt Clara" wrote:

wrote in message
.. .
In message ,
Gregory Blank wrote:

In article ,
wrote:

To answer you quite directly: for lack of using a better description
and to be concise.

If someone decides that "ISO 100 gives the best quality" and gets an
image that utilizes only 1/16th of the RAW values available, they would
have had a much better image if they had the camera set to ISO 1600
with
the same aperture and shutter speed. I have a hard time saying that
they "under-exposed" the image; it makes more sense to say that they
under-digitized it (quantized it) by using too low of an ISO.

Is it just me, or are you getting the ISO's wrong here? ISO 100 allows for
the greatest amount of exposure and equals the best possible image quality
(except for on film where you can go much lower).

.... only when a specific condition is met; that the "relative exposure"
or "exposure compensation" remains the same at the different ISOs, and
that the camera/lens combo has the shutter speeds and f-stops to
maintain the EC at all ISOs.

Of course, some of those captures will be blurred, or have less DOF than
desired, or poor, wide-open optics.

This is not a paradigm that most people operate under, but given a
limited amount of available light, the best capture is obtained by
making the photographic compromises between f-stop, shutter speed, vs
the S/N ratio in the sensor (analog), and using the ISO that uses the
biggest part of the range of RAW capture values (the highest ISO) that
doesn't clip desired highlight detail.

The rule of "using the lowest ISO" for maximum capture quality only
works when DOF and shutter speed are not issues (static scene with
tripod and MLU).

People aren't being
mislead, it's exactly the same as in the film world.

Not exactly. The film process is analog. The digital process starts
out with an analog exposure, and then digitizes a sub-range of it. Poor
digitization at low ISOs is just as bad, and possibly worse, than
amplified sensor noise at high ISOs.

Faster films/iso's
means better low light/hand held photographs, but at the expense of
grain/noise.

That compromise is not always present with digital.

As I said in another post, if the brightest part of your scene is a
middle grey, and a majority of it, then shooting at +2 EC at 4x the ISO
as a "normal exposure" will result in less noise, not more, if the
camera is truly using 4x the amplification at 4x the ISO. Shooting at
+2 EC at the lowest ISO will be even better, noise-wise, but potentially
impractical in terms of f-stop and shutter speed.

--


John P Sheehy

David J Taylor wrote:

To get the higher ISOs, for example, and using simple numbers:

- in normal mode, the converter digitises a voltage range of 0..4V to a
digital range of 0..4000. This means it must be accurate to 0.001V. The
ADC will work by comparing a signal with a 4V reference value.

- in ISO 3200 mode, the signal range is now just 0..1V. To do this, you
could

either:

-- quadruple the values from the ADC, turning it into a device digitising
0..1V to a digital range of 0..1000, but multiplying each result by 4, so
that 0..1V returns digital values of 0..4000. The quantisation steps are
still 0.001V, but as the digital values are quadrupled, the digital levels
in the image will be in steps of 4; 0, 4, 8, 12, etc.

or:

-- reduce the reference voltage in the converter so that it measures the
analog voltage against a 1V reference, but still returns values 0..4000.
The quantisation steps are now 0.0025V. Whilst the analog accuracy of the
converter may not justify the full 0,0025V steps, digitising this way may
produce a slightly more accurate result than simply quadrupling the
values. The digital levels will still be in steps of 1.

Perhaps there is a simple "double the output of the converter" happening
in cameras where this quantisation is observed? Perhaps both methods are
used to get two extra ISO steps?

Yes, I owuld guess that it is so. When below, eg, ISO 1600, the ADC is
stepped, and at 1600 and above the sensor readings are left shifted.
There may or may not be an overlapped zone where some gain is in the ADC
and some gain is in shifting left.

Then, there may be some filtering in the ADC as well as (of course)
making the RGB image from the Bayer image that will disgusise (filter,
badly or otherwise) the quantization effects.

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: Remove FreeLunch.

wrote:

"Quantization" would be clearer, I suppose, even if this is correct. I
see the word posterized much more than I do quantized, nowadays, and
assumed the former was more in common usage. People almost always refer
to an image with too few color levels to represent smooth gradients as
"posterized".

That comes out of the appearance of the image. Photographers after all.

Quantization noise is further disguised in character, of course, by the
conversion from bayer to RGB.

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: Remove FreeLunch.

wrote in message
...
SNIP
The rule of "using the lowest ISO" for maximum capture quality
only works when DOF and shutter speed are not issues (static
scene with tripod and MLU).

True, but for me that's automatically solved when I "expose to the
right". If the shutterspeed and aperture are cast in concrete (which
they often aren't), then all that's left is to correct with the ISO
setting (not to influence the exposure meter, I'd probably use Manual
in such a situation) in order to change the amplification on the
analog signal before ADC.

However, if capturing the full scene Dynamic Range is important, I'd
probably choose ISO 100 if I need to get the best sensor DR output,
and adjust the Depth-of-Field / camera shake / subject motion
trade-off.

Bart

wrote in message
...
SNIP
This is what I'm talking about; this is what I think exposure really
means (for a given subject intensity, of course). It is often used,
however, for the relative brightness of an image converted with 0
exposure adjustment, which, IMO, is more appropriately called
digitization.

The problem is that you essentially talk about Manual exposure
setting, which changes the ISO setting in an analog signal gain
control. However, in all other (non-manual) exposure settings, the
time x aperture metering settings change in function of the ISO
setting, in addition to the gain control. The latter applies to the
majority of images shot, and thus ISO also changes the amount of
noise.

I'd suggest just using Exposure when using the EV, and EV + gain
control when shooting Manual.

Bart