Digital Sensor's sensitivity curve

[TorontoJim]

First, let me apologize. I am new to digital photography, but not (sigh) at all new to taking pictures. I've done a couple of weddings, sold a few prints, but all "by accident" as my "job" was not photographic.
That said, I have to question digital photography and its information.

• for example - in contrast -

Ilford FP4 was known to be (slower) and much milder in contrast than Ilford XP5. Both were good B&W films. The response curve of each was well understood. You could pick fast/harsh or slow/subtle.

• yet -

today we have sensors that boast extreme ISO rating ranges. If this were a simple matter of taking floating point values and changing the sensitivity range, OK. But I suspect it is not: higher ISO means higher incidence of noise. DPreview tests, and all our own experience, will confirm this feeling.

• I believe I have set up my question semi-adequately, so here goes: -

Why can we not have ISO versus noise graphs? Why do we not know the fundamental sensitivity range of the sensor? (Think of Ansel Adams "fudging" the exposure so it would be "inside" his film and/or his print paper.)
I think we have been over and under informed here.
more to the point,
What Do You Think on this issue:

• do you understand the sensitivity versus noise of your camera's sensor?
• do you understand what "noise reduction" means in your camera? at every light level?

To Put It Another Way

I can tell you, from memory, the contrast ratio possible in a print. In a slide. guess, on a monitor.
Who can tell me the contrast ratio possible on a sensor? memory card?
In your camera?
Should we be informed to know this?
hope this helps us all to ask, loudly.

 

[Chuck Gardner]

The camera's sensitiviity curve is very easy to determine. Just shoot a white or gray card that fills the entire frame and bracket your exposure from the point where the card is pure white until it is pure black.

Then look at the camera and Photoshop levels histograms for the shots and their RGB values.

The card, because it is a single tone will produce a very narrow spike on the histogram. The histogram represents the dynamic range of the camera. As exposure is reduced from the point the card is reproduced as white the spike will march left across the histogram. If exposure is bracketed in one-stop intervals it will look something like this:



That is the result of the test I did with my Canon 20D. The red lines indicate where on the camera histogram the spikes occurred. The small patches of tone are cut and pasted from the file with that histogram position. The number in the tonal swatch is the RGB value for the tone. It is a road map to both the sensitivity of the camera and how many stops of detail it can accurately record, and the actual tonal values each spot on the horizontal scale of the histogram represents on the print.

Once you get a bit more experience with digital you will find that the noise issue is a bit like dealing with film grain. Its really not noticed much at all at lower ISO speeds. Up around 400 you need to make a decision to either add artificial light to keep the ISO and noise down below the level where it affects image quality, or stick with ambient light and live with the grain. Lighting is so situational there will never be one best answer for every situation. Experience, not signal-to-noise graphs, will be the best guide.

The illustration above is from my histogram tutorial:
http://super.nova.org/DPR/ZoneSystem/Histogram.pdf
CG

 

[John Paul]

The camera's sensitiviity curve is very easy to determine. Just
shoot a white or gray card that fills the entire frame and bracket
your exposure from the point where the card is pure white until it
is pure black.

Then look at the camera and Photoshop levels histograms for the
shots and their RGB values.

The card, because it is a single tone will produce a very narrow
spike on the histogram. The histogram represents the dynamic range
of the camera. As exposure is reduced from the point the card is
reproduced as white the spike will march left across the histogram.
If exposure is bracketed in one-stop intervals it will look
something like this:



That is the result of the test I did with my Canon 20D. The red
lines indicate where on the camera histogram the spikes occurred.
The small patches of tone are cut and pasted from the file with
that histogram position. The number in the tonal swatch is the RGB
value for the tone. It is a road map to both the sensitivity of
the camera and how many stops of detail it can accurately record,
and the actual tonal values each spot on the horizontal scale of
the histogram represents on the print.

Once you get a bit more experience with digital you will find that
the noise issue is a bit like dealing with film grain. Its really
not noticed much at all at lower ISO speeds. Up around 400 you
need to make a decision to either add artificial light to keep the
ISO and noise down below the level where it affects image quality,
or stick with ambient light and live with the grain. Lighting is
so situational there will never be one best answer for every
situation. Experience, not signal-to-noise graphs, will be the
best guide.

The illustration above is from my histogram tutorial:
http://super.nova.org/DPR/ZoneSystem/Histogram.pdf
CG

I did a similar test last fall when I was considering upgrading from my trusty Minolta A1 to an Oly E330 (I now have an E510 on order). But I did one big difference from what Chuck suggests, I used 2 cards with about a 1 stop difference. (The white and gray sides of cardboard.) I also shot a range of exposures from the point it looked completely washed out to total black. My analysis included looking to see how long I could tell the 2 cards apart.

Straight out of camera I was surprised to find that the A1, with a sensor 1/4 as big, had about a 1/2 stop advantage at the dark end. But then I started pulling up the shadows in PS (levels works if you want to test the dark end and don't care about the overall picture), and found a very different story. First, I hadn't extended my exposures far enough. Even with rgb values around 10, there was still the potential to pull up detail, depending on the camera. But with the A1, any difference which might have been there was soon overwhelmed by noise. On the other hand, the Oly was still able to produce an easily recognizable difference as far down as I had done exposures.

I think that Phil gives us something comparable with his dynamic range tests on dslr. The out-of-camera jpg measure is mostly a test of how the designers decided to render the sensor response. But his "best that could be done with ACR" curve seems similar to what I did in stretching the dark end. He usually comes up with several stops more DR than the jpg shows. If only he would do such a test on other grades of cameras. My question in thinking about going from my superzoom A1 to a dslr was mostly about increased DR, but Phil's testing doesn't show comparisons across classes. I grant that DR differences may not be a big issue within the cheaper classes, but it would be nice to have his objective testing for a least a few representative samples of those classes so we have an idea of what we stand to gain when we upgrade.
--
John Paul

-- changing the subject line makes threads easier to follow --

 

[John Paul]

The "curves" function in Photoshop (which I think they finally added to the most recent version or two of PS Elements), may give us control over the kind of effect you were citing for the different films. There is the issue of the native DR of a given sensor at a given ISO, but beyond that we have a lot of choice over how much contrast we get. (And there is my favorite trick: try sharpening with a huge radius, 50 or more. It serves as a regional contrast increase.)
--
John Paul

-- changing the subject line makes threads easier to follow --

 

[Svein Eriksen]

What about the graphs at the bottom of this page?
http://www.dpreview.com/reviews/panasonicdmcl1/page16.asp

 

[Chuck Gardner]

One of the more informative things I did photographically was to start by learning the Adams zone system by reading his books and then a few years later work in the National Geographic Photomechanical lab producing halftones, duo-tones and tri-tones used for the publications and maps. That was back in the mid-1970s when the process was entirely conventional.

There is actually a significant technical difference between how Adams manipulated contrast by aligning the end points to the scene illumination range and how the conventional halftone process manipulated the midpoint to change contrast. But the net effect to the eye is about the same because the eye is easily fooled into thinking that a manipulation of the mid tone placement relative to the highlights is actually a net increase in scene contrast. The levels function in Photoshop with its midtone slider is modeled on the way the halftone process could fool the eye by changing the tonal relationship between the mid-tone values and the two fixed end points.

For example, when you open a file in Photoshop and move the middle slider to make the "shadows" lighter you are really only changing the internal contrast relationship of the recorded information, not increasing the overall contrast of what was record the way Adams did during the capture via negative development to match scene range and negative density range to the print tonal range.

I've long felt that Adams real genius was being one of the first to recognize the role human perception has on the holistic process of seeing a scene and capturing its emotional impact on a piece of silver coated paper. What is outstanding about a photo like Half Dome is the fact we can actually see a greater range in the print than in person looking at it because of the way Adams manipulated the capture. In person the pupils would constrict and the shadow detail just wouldn't be seen. Adams, by measuring the illumination range and tailoring it to range of the print via negative development allows us to see everything including the nuances of shadow detail our eyes and brains wouldn't be able to take in all at once. He even took it a step further, realizing that when we looked at the print the normally bright sky would have a similar negative effect when viewing the shadow detail of the print. So he used a red filter on the camera to darken the sky much darker and it would appear by eye, In fact filters and his knowledge of how they would affect the tonal balance and perception of tone in the final print where a key element in many of his captures.

We can't duplicate what Adams did with B&W film with digital because is impossible to expand the range of DETAIL in the scene the sensor can capture with a single exposure beyond about 6-7 stops. But thats really not a problem because its about the same limitation that color negative films have with respect to recording the scene and is more tonal range than can be rendered with transparency films.

Digital will likely never be able to capture a 12 stop range of illumination as Adams did with a B&W negative but the goal of reproduction shouldn't be to exactly reproduce the illumination range but rather to create a facsimile which is believable enough to evoke a similar visual response and emotional reaction in the mind of the viewer. Just stop for a minute and consider that the world isn't B&W and you'll be halfway towards a better understand of how best to utilize the tools we currently have.

While we can't duplicate Adams approach with digital, we can adopt his approach of understanding and exploiting the recording medium to its limits and understanding how those limits fit within the broader context of the way the eyes and brain interpret manipulation of the tonal values. The most important lesson I've learned after being in the facsimile reproduction business for 35 years is that if the "anchor" values of black and white correctly the viewer will use them as a frame of reference and accept everything in the middle as being an accurate rendering of the scene. But if the highlights are blown or underexposed and the darkest shadows are gray not black in a scene which a person's stored memories say should be black the image will be perceived as being not quite right. Once the two end points right, then the perception of shadow detail vs highlight contrast can be manipulate with either levels or curves.

With digital getting the highlights right is of paramount importance because its the highlight detail and subtle variation in the highlight-to-midtone range which most greatly influences how the image is perceived. The only real requirement on the shadow end is that the shadow be deep and black. With digital that is seldom a problem. Yes shadow detail falls where it may and some of it is lost, but in most cases the primary function of the dark stuff in a photo is simply to push the eye of the viewer towards the more important message in the highlights. Contrast is what leads the eye in a photo. The greater the contrast the stronger the attraction is to what contrast the most with the background.

It's possible to expand the range the camera captures with the layering of multiple exposures (DHR) but that's a bit of a double edge sword. First because in photos where center of interest is a lighter object on a overall dark field the overall goal and objective is to lead the eye of the viewer to it. It's the progression on tonal values from dark to light on a dark background or light to dark on a light one that leads the eye. Too much shadow detail can actually create a distraction and dilute the overall message of the photo. The more the range is expanded to capture every subtle nuance of tone in a scene via HDR the more and more everything starts to look the same.

CG

 

[DavidMillier]

Hi

Your post makes it sound as though these matters are not well understand and properly described which isn't the case.

First, let me apologize. I am new to digital photography, but not
(sigh) at all new to taking pictures. I've done a couple of
weddings, sold a few prints, but all "by accident" as my "job" was
not photographic.
That said, I have to question digital photography and its information.

• for example - in contrast -

Ilford FP4 was known to be (slower) and much milder in contrast
than Ilford XP5. Both were good B&W films. The response curve of
each was well understood. You could pick fast/harsh or slow/subtle.

With film, the response curve was the result of film speed and development choice not just film type.

• yet -

today we have sensors that boast extreme ISO rating ranges.

Film speed was a somewhat variable number and depended on development. Adams recommended testing to determine a personalised film speed.

The senstivity of electronic sensors is fixed. Higher ISOs on digital cameras are really "pushing" through the use of electronic amplification. Higher iso settings reduce the max quality available at the base or native ISO.

If this
were a simple matter of taking floating point values and changing
the sensitivity range, OK. But I suspect it is not: higher ISO
means higher incidence of noise. DPreview tests, and all our own
experience, will confirm this feeling.

• I believe I have set up my question semi-adequately, so here goes: -

Phil's noise patches are effectively this.

Why can we not have ISO versus noise graphs?

Phil's dynamic range tests are effectively this...

...but note that sensor dynamic range involves a degree of subjectiveness because DR is limited by noise and any tester needs to make their own judgment about what constitutes acceptable noise. Kodak got into trouble here with the 14n: it displays a wider DR than most cameras at the cost of severe and unpleasant shadow noise.

Why do we not know the
fundamental sensitivity range of the sensor? (Think of Ansel Adams
"fudging" the exposure so it would be "inside" his film and/or his
print paper.)

I think we have been over and under informed here.
more to the point,
What Do You Think on this issue:

• do you understand the sensitivity versus noise of your camera's

sensor?

• do you understand what "noise reduction" means in your camera? at

every light level?
To Put It Another Way

The contrast ratio of sensors using Phil's test scheme appears to cluster around 8 to 8.5 stops.

Memory cards are digital data storage devices that store information as coded numbers - they don't and can't have a contrast range

I can tell you, from memory, the contrast ratio possible in a
print. In a slide. guess, on a monitor.
Who can tell me the contrast ratio possible on a sensor? memory card?
In your camera?
Should we be informed to know this?
hope this helps us all to ask, loudly.

--
Galleries and website: http://www.whisperingcat.co.uk/mainindex.htm

 

[BJL]

Firstly, review sites like this one do produce graphs of noise levels vs ISO speed setting, even if manufacturers do not.

Secondly, the ISO speed of a sensor that best matches the ISO speed of a film is the one where shadow noise levels give about three of four usable stops below the mid-tones. With DSLR's that is usually well above the minimum usable ISO speed, which is instead based only on the ability to avoid blown highlights.

(The ISO definition of film speed is based in part of placing midtones about four stops above "film base plus fog", the rough counterpart of sensor dark noise.)

My rough reading is that the "film equivalent" ISO speed of current DSLR sensors mostly ranges from 800 to 1600, with maybe the new 1DMkIII going as high as 3200 and the (now discontinued?) Olympus E-500 as low as 400.

In some sense, a lot of the wide ISO setting range of DSLR's is because their sensor response curves have a straight line section far wider than with film (but with no shoulder above!), and so one can adjust Exposure Index (ISO setting) considerably while still getting a good three stops or so above and below the midtones, good enough for scenes of normal subject brightness range.

Think of a DSLR's lower ISO settings as "pull processing", with the associated abundance of shadow range.

--

I want all my lenses to be f/4 or brighter and adequately affordable and portable.

Higher usable ISO from bigger pixels is useless if it forces me to use longer, slower telephoto lenses to get the same resolution.

 

[foot]

Well I'm very much a newbie, so take all this with a big grain of salt.

Here goes:

1) With film, you kept the camera body, and changed the film. Each type of film had it's characteristics, and could be used to achieve different pictures of the same scene. In digital, each body has an unchangeable sensor, which fixes the physical properties of the camera.

Currently there are several different sensor types: ccd, cmos, foveon, and the fuji sensor. Each has their following. For example, wedding photographers like the fuji sensor, since it has special electronics on the sensor chip to give a larger dynamic range.

On top of the sensor, there are the micro-lenses and anti-aliasing filters. These also are fixed and can't be change. This is why different camera manufacturers can use the same sensor (for example, the sony CCD sensor, which is used by sony, nikon, pentax, etc) and get different results.

Each of the physical pixels (for example red, green, blue) has different sensitivity and noise curves.

And the chroma and luminance channels have their own sensitivity and noise curves. Some cameras have more chroma noise, and some have more luminance noise.

And different types of noise are easier to do Noise Reduction on than other types.

So after the data is read from the sensor chip, a certain amount of proprietary signal processing is done to get the "RAW" data.

At this point, many photographers stop the in-camera processing, and store the "RAW" data onto the memory card. They do any further data processing on their computer, using photoshop, etc (ie: post-processing).

The RAW data will have more information per color channel than the jpeg - usually 12, 14, or 16 bits vs the 8-bits in the jpeg file.

Others prefer to let the camera do the processing of RAW data into jpeg.

Since the manufacture has proprietary information, the camera can often do a better job than most ppl.

Since jpeg is a lossy compression, many compromises were made when the file format was designed. Not all color channels are treated the same!

Also the "white balance" comes into play. It's much harder to change the white balance from a jpeg file than a RAW file. Often much more noise is created as a side-effect. With the RAW file, you don't have this limitation.

Some cameras let you select/download the "color curvers". With color curves, you can get much different jpeg images with the same RAW data.

Well sorry I went on so long! Here's the walk-away:

camera sensor capabilities + color curves = film characteristics

PS: since I'm a newbie, anybody have corrections or feedback, well I'm here to learn! and Thanks!

First, let me apologize. I am new to digital photography, but not
(sigh) at all new to taking pictures. I've done a couple of
weddings, sold a few prints, but all "by accident" as my "job" was
not photographic.
That said, I have to question digital photography and its information.

• for example - in contrast -

Ilford FP4 was known to be (slower) and much milder in contrast
than Ilford XP5. Both were good B&W films. The response curve of
each was well understood. You could pick fast/harsh or slow/subtle.

• yet -

today we have sensors that boast extreme ISO rating ranges. If this
were a simple matter of taking floating point values and changing
the sensitivity range, OK. But I suspect it is not: higher ISO
means higher incidence of noise. DPreview tests, and all our own
experience, will confirm this feeling.

• I believe I have set up my question semi-adequately, so here goes: -

Why can we not have ISO versus noise graphs? Why do we not know the
fundamental sensitivity range of the sensor? (Think of Ansel Adams
"fudging" the exposure so it would be "inside" his film and/or his
print paper.)
I think we have been over and under informed here.
more to the point,
What Do You Think on this issue:

• do you understand the sensitivity versus noise of your camera's

sensor?

• do you understand what "noise reduction" means in your camera? at

every light level?
To Put It Another Way
I can tell you, from memory, the contrast ratio possible in a
print. In a slide. guess, on a monitor.
Who can tell me the contrast ratio possible on a sensor? memory card?
In your camera?
Should we be informed to know this?
hope this helps us all to ask, loudly.

 

[jeffkrol]

best digital sensors, 40:1 and pretty darn linear till messed with..you can apply any curve you want.
http://www.ronbigelow.com/articles/raw2/raw2.htm
http://www.betterphoto.com/article.asp?id=50
How Kodak measures base ISO.. based on S/N ratio and sensor saturation levels..
http://www.qualiteitems.com/images/ISO.pdfi
--
360 minutes from the prime meridian. (-5375min, 3.55sec) 1093' above sea level.

'The exposure meter is calibrated to some clearly defined standards and the user needs to adjust his working method and his subject matter to these values. It does not help to suppose all kinds of assumptions that do not exist.'
Erwin Puts

 

[John Sheehy]

That is the result of the test I did with my Canon 20D. The red
lines indicate where on the camera histogram the spikes occurred.

The results you got are about the RAW conversion, not the camera per se.

--
John

 

[alanr0]

best digital sensors, 40:1 and pretty darn linear till messed
with..

Not sure what this is based on. Depending on your quality criteria are, current DSLRs have between 6 and 12 stops dynamic range - a factor of 4000:1 in intensity. Most will show a pretty linear response over an intensity range of at least 1000:1.
http://www.imaging-resource.com/PRODS/K10D/K10DIMATEST.HTM
http://www.clarkvision.com/imagedetail/dynamicrange/

you can apply any curve you want.
http://www.ronbigelow.com/articles/raw2/raw2.htm

The numerical examples for JPEG are somewhat misleading. A sensible JPEG engine will apply a tone curve (usually close to SRGB) before converting from a 12 bit to an 8 bit representation. There is loss of tonality, but it is not as bad as he makes out.

Cheers.
--
Alan Robinson

 

[BJL]

The words "Signal to Noise Ratio" get uses to describe two very different things, with 40:1 is relevant to one, 1000:1 to the other.

One is local, the ratio of signal to noise in the output of any individual photosite or pixel.

A Kodak document says that a 40:1 S/N at a pixel is visually "excellent", while 10:1 is a bare satisfactory level.

The other is global, related to dynamic range: the ratio of the maximum signal level that a photosite can detect without being blown out to the "dark noise" level, the minimum noise level present at any pixel, even ones in complete darkness.

This is what needs to be about 1000:1 for decent dynamic range, corresponding to about what 10 bit A/D convertor can handle. 4000:1 matches the abilities of 12-bit A/D, and is about as high as almost any current sensor give AFAIK.

--

I want all my lenses to be f/4 or brighter (for good AF speed) and adequately affordable and portable.

Higher usable ISO from bigger pixels is useless if it forces me to use longer, slower telephoto lenses to get the same resolution.

 

[alanr0]

The words "Signal to Noise Ratio" get uses to describe two very
different things, with 40:1 is relevant to one, 1000:1 to the other.

I did not mention 'Signal to Noise Ratio', and neither did jeffkrol. I was talking about dynamic range, and pointing out that the references he linked provided rather questionable information.

One is local, the ratio of signal to noise in the output of any
individual photosite or pixel.
A Kodak document says that a 40:1 S/N at a pixel is visually
"excellent", while 10:1 is a bare satisfactory level.

I am familiar with the ISO 12232 definitions of 'first acceptable' and 'first excellent' SNR. They are used to determine how far the exposure index can be reduced (i.e. how far the ISO can be increased) while retaining an acceptable or excellent image quality.

In a typical measurement the 40:1 is not at 'an individual pixel'. According to the standard, it is measured from the standard deviation over a 64x64 pixel square at the average exposure level.

As far as I can tell, this was not the meaning jeffkrol used when he stated "best digital sensors, 40:1". I took this to mean that the best digital sensors could achieve an SNR of only 40:1, which is a gross under-estimate.

This certainly seems to be in the meaning in the article he linked to:
http://www.betterphoto.com/article.asp?id=50

From the article: "The human eye can perceive a contrast ratio of 800:1; the best slide films, 30:1; and the best digital sensors, 40:1." This is rubbish.

The other is global, related to dynamic range: the ratio of the
maximum signal level that a photosite can detect without being
blown out to the "dark noise" level, the minimum noise level
present at any pixel, even ones in complete darkness.
This is what needs to be about 1000:1 for decent dynamic range,
corresponding to about what 10 bit A/D convertor can handle. 4000:1
matches the abilities of 12-bit A/D, and is about as high as almost
any current sensor give AFAIK.

Yes. I was indeed talking about dynamic range. I assumed this was what the OP meant by 'contrast ratio'.

Cheers.
--
Alan Robinson

 

[John Sheehy]

This is what needs to be about 1000:1 for decent dynamic range,
corresponding to about what 10 bit A/D convertor can handle. 4000:1
matches the abilities of 12-bit A/D, and is about as high as almost
any current sensor give AFAIK.

Noise is a bigger issue than bit depth, with current cameras. The only consumer camera I know of that has noise low enough fo0r bit depth to contribute a relevant amount of noise is the Pentax K10D at ISO 100, with a noise "floor" (I hate the possible connotations of that term) of 0.9 ADU/DN. Next behind that are the Canon 1-series and Fuji DSLRs (at ISO 100 also), at about 1.3, and the rest of the DSLRs are mainly in the 1.7 to 4.0 range at ISO 100 (some tiny sensor cameras are also in this range).

--
John

 

[BJL]

Alan, I agree, and reply mainly to clarify my comments, and thank you for some information.

The words "Signal to Noise Ratio" get uses to describe two very different things, with 40:1 is relevant to one, 1000:1 to the other.

I did not mention 'Signal to Noise Ratio', and neither did jeffkrol. I was talking about dynamic range ...

Agreed: take my comment as an explanation of why that 40:1 figure is not relevant to dynamic range needs, but refers to something else, despite the confusing use of S/N ratio for both, which probably led to the misinterpretation of that 40:1 figure. I am all for your 1000:1 (and up) DR criterion!

... and pointing out that the references he linked provided rather questionable information.

Perhaps not "questionable" if interpreted correctly, but misinterpreted when taken as a measure of DR.

I am familiar with the ISO 12232 definitions of 'first acceptable' and 'first excellent' SNR.

Great: can you give a reference for the details? (One that does not involve sending a big check to ISO!) I have read about the ISO 12232 definitions only through brief references to them in document at Kodak's web site, and I would love to see the full formal definitions.

In a typical measurement the 40:1 is not at 'an individual pixel'. According to the standard, it is measured from the standard deviation over a 64x64 pixel square at the average exposure level.

Agreed: I spoke loosely and of course one cannot measure a random variation without a sufficiently large sample from which to compute a meaningful sample variance. Thanks for the details on ISO procedure for that measurement.

What I meant by "per pixel" is only that it is the ratio of signal level at photosites receiving a particular level of illumination to the (estimated) noise level at pixels at that same level of illumination, as opposed to DR measurement where it is, loosely, the ratio of signal at highlight pixels to noise at shadow pixels.

From the article: "The human eye can perceive a contrast ratio of 800:1; the best slide films, 30:1; and the best digital sensors, 40:1." This is rubbish.

Agreed: total rubbish!

--

I want all my lenses to be f/4 or brighter (for good AF speed) and adequately affordable and portable.

Higher usable ISO from bigger pixels is useless if it forces me to use longer, slower telephoto lenses to get the same resolution.

 

[BJL]

John, I agree, and thanks for the details. They confirm the hidden agenda of my comments on bit depth, which is that people who clamor for higher bit depth A/D convertors, and think for example that there would be an advantage to going beyond about 12 bit A/D with current sensors are barking up the wrong tree. Indeed, it would be very strange for it to be otherwise: increasing the bit depth of of an A/D convertor is surely far easier and less expensive that increasing sensor DR, so the minor cost saving of not having enough bit depth to get the most out of the signal from the sensor would be an absurd engineering decision; a "false economy".

This is what needs to be about 1000:1 for decent dynamic range, corresponding to about what 10 bit A/D convertor can handle. 4000:1 matches the abilities of 12-bit A/D, and is about as high as almost any current sensor give AFAIK.

Noise is a bigger issue than bit depth, with current cameras.

The only consumer camera I know of that has noise low enough fo0r bit depth to contribute a relevant amount of noise is the Pentax K10D at ISO 100, with a noise "floor"... of 0.9 ADU/DN.

It is interesting that this particular camera is the leader. What is your source for these measurements? I am curious to see how the new Panasonic 10MP 4/3" nMOS sensor with 4.7 micron pixel pitch rates.
--

I want all my lenses to be f/4 or brighter (for good AF speed) and adequately affordable and portable.

Higher usable ISO from bigger pixels is useless if it forces me to use longer, slower telephoto lenses to get the same resolution.

 

[alanr0]

I am familiar with the ISO 12232 definitions of 'first acceptable' and 'first excellent' SNR.

Great: can you give a reference for the details? (One that does not
involve sending a big check to ISO!) I have read about the ISO
12232 definitions only through brief references to them in document
at Kodak's web site, and I would love to see the full formal
definitions.

Like you, I was interested, but not keen to come across with the Swiss Francs.
There is a pre-publication draft of the 1998 standard here:
http://www.opsci.com/AppNotes/ShortOAN-004ISORatingDigitalCameras.pdf

By default you only get the first few pages, but you can download a full version if you first register on http://www.opsci.com/

I believe it is reasonably close to the published 1998 version. I have not read the latest version of the standard (updated 2006)

http://www.iso.org/iso/en/CatalogueDetailPage.CatalogueDetail?CSNUMBER=37777&ICS1=37&ICS2=40&ICS3=10&scopelist=

More discussion here:
http://forums.dpreview.com/forums/read.asp?forum=1018&message=20483785

Cheers.
--
Alan Robinson

 

[ppage]

A really excellent post. Thank you.
Peter

 

[BobTrips]

From the article: "The human eye can perceive a contrast ratio of 800:1;
the best slide films, 30:1; and the best digital sensors, 40:1." This is
rubbish."

I'm not suggesting that the correct number is 800, I have no idea what the actual number might be, but....

The human eye when fully dark adapted (about 30 minutes in full darkness) is pretty sensitive. And when allowed to adapt to bright light for a few minutes can operate in very bright conditions.

This "total range" might well greatly exceed any photographic medium.

--
bob

Sleepers
http://picasaweb.google.com/Bobfwall/Sleepers

The Blind Pig Guild - A Photo/Travel Club
http://www.jeber.com/Clubs/Blind-Pig/

Travel Galleries
http://www.pbase.com/bobtrips