"Exposure" vs "Digitization

David J Taylor wrote:

While we are talking about terms, I find JPS's use of the term
"posterisation" confusing at best, and meaningless in a signal (or image)
processing context. I think what he means may be "quantisation", the fact
that an infinite range of analog values must be represented by a limited
range of digital levels. Normally, sufficient digital levels are
available and it is the accuracy of the analog signal which determines the
signal-to-noise ratio of the system. However, if the quantisation steps
are too large, it becomes the quantisation process itself which limits the
signal-to-noise ratio.

Yes, when a photo-geek starts talking about "posterization", they are
referring to what anyone with a background in signal processing would
call "quantization noise".

http://en.wikipedia.org/wiki/Quantization_noise

Many fields have a maddening "terminological thicket" to penetrate...

In article ,
wrote:

Thats a rather extreme example,... & it seems unlikely.

If you think that's unlikely, you haven't been reading people's posts,
or DPReview. The problem of people under-digitizing at ISO 100 is
epidemic, because of the myth that ISO settings cause noise.

Up until this moment:

The only posts I have read are yours and mine regarding this
thread. Hey under exposure is under exposure. On film you get nothing.


"Why is the sky so noisy in my ISO 100 picture" is a common question.
Of course, it is not just sensor-noisy, it's also highly posterized as
well, and would have looked better if taken at a higher ISO setting,
even with the same aperture and shutter speed. If they were using a
tripod, of course, they could have had a good digitization at a higher
absolute exposure at a lower ISO. I personally don't use ISO 100 very
often, but aim for ISO 200 if I can do it with a full digitization.
Blooming looms just above RAW value 4095 at ISO 100. In my experiments,
the trade-off between sterile posterization and noise indicates that
there is very little value in using ISO 100 over ISO 200 on my Canon
20D. The shadows are of approximately equal worth.

That is: is it better for noise, range and color? Or does one make the
choice to keep two and drop one from the equation? Because if its better
for all three you would only need one ISO setting and not any
supplemental light sources.

You add flash/or lights as needed to make the Iso 100 image.

Not in available light photography, you don't.

So knowledge of basic photography is a good thing if one wants to make
good pictures.

Digital
does not solve the problems that exist beyond the scope of the camera-
lighting. & more likely It never will. Lighting is separate set of
issues and require knowledge.

Optimal lighting is different for digital and film.

To a degree maybe, but fairly close for slide film and the sensor.

Color film
generally wants to see sunlight or tungsten, depending on the film.

Most digitals have neither sunlight nor tungsten as their native white
balance. The native balances generally run from magenta to pink
lighting with RGB bayer cameras. My Canon DSLRs get the best images
with lighting that is a stop stronger red than green, and a half stop
stronger blue than green.

Like I stated prior it relative to what the maker imparts.

I can't seem to state this enough to
people, its something schools should teach

Should be taught specific to digital, but I doubt that there are many
teachjers who know the difference.

Should be taught specific to what works it incorporates many
discipline fields. Video, film and digital the principles of light
adjustment should be considered a portion of many curriculum including
architectural design,...they maybe in that but I am just stating that
its an important branch of making pictures.

Available-light photography can only be improved by maximizing exposure
without clipping, or using filters over the lens if there is enough
light.

It all boils down to understanding, that is knowing when to use that
filter and when not,...its the same for film images.

--
LF Website @ http://members.verizon.net/~gregoryblank

"To announce that there must be no criticism of the President,
or that we are to stand by the President, right or wrong,
is not only unpatriotic and servile, but is morally treasonable
to the American public."--Theodore Roosevelt, May 7, 1918

wrote:

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.

The thing is, you're the only person I have *ever* seen talk in terms of
"absolute exposure", or compare different ISOs at the same aperture and
shutter speed. I'm not saying that's not a valid way to think about it,
I just don't see how it could be useful to me. ISO 100 *does* give the
best quality, as long as you don't underexpose it, and it's a given that
a proper exposure at an elevated ISO rating is better than underexposing
at ISO 100.

Now, there is another angle to the whole thing, and that is the 12-bit
A/D conversion. We've discussed previously how the dynamic range of
current sensors is not limited by the sensor's capability, but rather
by the A/D conversion. This presents an interesting situation. Imagine
that a camera used 16-bit A/D conversion. Imagine that the extra range
actually *did* use all of the data available from the sensor. You now
have a situation where higher ISO settings are meaningless, and the
camera would have to be marketed as (for example) ISO 100 with *no*
higher settings. Imagine the outcry!

The simple fact that higher ISO settings exist and are useful tells
us that the A/D conversion is not using all of the data the sensor is
providing. Higher ISOs are accomplished by amplifying the signal.
If you can usefully amplify the signal to ISO 800, that means there
was a signal there in the first place to amplify, one that *could*
have been used at ISO 100, but was ignored at that setting. If no
data from the sensor were ignored, there would be nothing left to
amplify, and ISO 200 would just be ISO 100 with one stop less of
range and no actual advantage whatsoever. That is, it *would* be
better to underexpose at ISO 100 and then push it in processing, to
avoid the amplification step.

So, it seems that either 16-bit A/D conversion is more complicated
to put into a camera than it sounds, or we are having our dynamic
range artificially limited in order to allow camera manufacturers
to say that their cameras can go to ISO 800 or whatever.

Of course, if the sensor can provide more range than a 16-bit
conversion would need, then there would still be room for higher
ISO settings.

--
Jeremy |

wrote:


I didn't say that the term exposure doesn't apply at all. I said that
it wasn't a good way to describe the relative digitization at an ISO
setting.

And most photographers will say: "So what? Now let's meter the scene
and set the exposure."


You may be right about the term "digitization" but you draw strange
looks becasue it is not the famillar term. And there really is nothing
wrong with the term exposure. That word says is all: Time X aperture.


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.

In engineering terms, sure. But few dedicated photographers are
engineers. And most engineers who photograph as a hobby or interest do
so as a pastime, not as an extension of their engineering life. Some
nerds excepted (not all engineers are nerds, far from it in my experience).

As to the 0 exposure adjustment, there is none, really. None of the
sensors have an ISO 134.6767 (whatever) setting where they are unity
gain. Few of the cameras record an ISO 100 (or other) sensitivity the
same as their competitors (or even models from the same co.)

You're raising a total non issue. Or at best an issue that is
meaningful to you and very few others. Photographers make photography
and think in photographic terms.

I know engineers who paint as well, but they don't talk about "pigment
carrying linum usitatissimum oils". They talk about oil paints.

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.

David J Taylor wrote:

An example from audio might be that when sounds are digitised to 8-bits
rather than 16-bits, there is an added roughness to the sound. An error
is introduced which depends on signal level. In an image, too few bits
would show as a contouring effect where what should be a smooth transition
of brightness levels instead appears as a finite series of perceptibly
different brightness bands.

I would call this effect quantisation errors, or more specifically errors
due to using too few bits to represent the signal. Is this what you mean
by posterisation?

I brought up the issue of quantization noise many months (over a year?)
ago on rpd. It seemed to sail over the heads of just about everyone. I
too did some signal processing work 10 or so years ago synthesizing
complex radar wave forms in realtime. Quantization noise was not a
problem in the synthesis (16 bit DAC) but for the system under test with
a very high dyncamic range, it was a serious issue when the SNR was very
low.

For photography, the quantization noise is the noise we typically see at
high ISO settings in the shaddow areas of the image. Some liken this
(erroneously) to film grain. However film grain has dimension across
the image (x,y), as well as in color error (z), whereas quantization
noise is dynamic (z) (color) only in digital cameras.

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:

The analog to slide film exposure is actually the analog exposure on the
sensor; the ISO settings of the digital camera are like setting different
ranges of exposure in a slide to be digitized by a scanner.

Why then, do we call utilizing the specified range "exposure".

Because when taking a picture, the sensor is exposed to light. When not
taking the picture, the sensor isn't exposed to light, particularly when
it comes to DSLRs.

Alan Browne wrote:
[]
I brought up the issue of quantization noise many months (over a
year?) ago on rpd. It seemed to sail over the heads of just about
everyone. I too did some signal processing work 10 or so years ago
synthesizing complex radar wave forms in realtime. Quantization
noise was not a problem in the synthesis (16 bit DAC) but for the
system under test with a very high dyncamic range, it was a serious
issue when the SNR was very low.

For photography, the quantization noise is the noise we typically see
at high ISO settings in the shaddow areas of the image. Some liken
this (erroneously) to film grain. However film grain has dimension
across the image (x,y), as well as in color error (z), whereas
quantization noise is dynamic (z) (color) only in digital cameras.

Cheers,
Alan.

Thanks, Alan.

I would be surprised if quantization noise were an issue at low SNR (i.e.
high ISO settings) in a digital camera, though, as the signal is amplified
before the ADC, so that the photon noise should swamp the quantisation
noise. However, I haven't sat down and done the sums....

Cheers,
David

In message ,
"David J Taylor"

wrote:


wrote:
[]
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.

In signal processing we might use the term "headroom". If the sound level
a system could capture was 8, and the loudest sound to be recorded was 4,
then we might say the headroom was 6dB (an engineering term for a factor
of 2 in linear voltage or current terms). Similarly, if the maximum value
from your image sensor is 4095, and the white level of a particular image
were 2047, then you also have a factor of two headroom. Were any part of
the image to exceed the maximum value - specular highlight for example -
we would say the value was clipped.

"Digitisation" is simply the process of converting analog to digital.

Implicit in everything I say here about digitization is the quality
factor; "well-digitized", "poorly digitized", etc.

A RAW file with the highest value as 600 in a specular highlight is
poorly digitized. A maximum of 2048 is not as well-digitized as a
maximum of 3800.
--


John P Sheehy

In message ,
"David J Taylor"

wrote:

While we are talking about terms, I find JPS's use of the term
"posterisation" confusing at best, and meaningless in a signal (or image)
processing context.

I've understood "quantization" to be a specific kind of "posterization";
one where the values are integers or multiples of integers, whereas
posterization can include things like 1 2 4 5 7 8 10 11, etc.

"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".
--


John P Sheehy

wrote:
In message ,
"David J Taylor"

wrote:

While we are talking about terms, I find JPS's use of the term
"posterisation" confusing at best, and meaningless in a signal (or
image) processing context.

I've understood "quantization" to be a specific kind of
"posterization"; one where the values are integers or multiples of
integers, whereas posterization can include things like 1 2 4 5 7 8
10 11, etc.

That's probably fair comment - most quantisers indeed use linear steps.

"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".

Yes, sometimes I see this effect used in television for, er, "artistic"
purposes. I've never understood why, personally! Were, instead, I
referring to something where this was a problem, I would simply say there
were too few quantisation levels, or that the quantisation step was too
large. Different steps in each of the RGB channels can even introduce
colour casts.

David