ISO - quality considerations

[JerryReed]

What is your experience with utilizing an ISO of 50 versus ISO 100?  The application that I am considering this for is table top still life, where the subject might be lit with just one light, providing for a wide-range of luminance.

Assume that the same f-stop will be used and only the shutter speed will be reduced to equalize the exposure.  In terms of dynamic range, would the use of a lower ISO, increase the OBSERVABLE DR at 20 by 24 inch prints from a 33 mega-pixel back? (SINAR 75LV).

I am interested also, in using the responses to see if the same presumed positive effect might be seen with my Canon 1Ds Mark II, which I just discovered is capable of utilizing a ISO 50.  Is there any reason not to expect that the presumed favorable results of utilizing a 50 ISO on the DB might be seen when ISO 50 is used on a sensor half the size?

Jerry Reed

[Bart_van_der_Wolf]

What is your experience with utilizing an ISO of 50 versus ISO 100?  The application that I am considering this for is table top still life, where the subject might be lit with just one light, providing for a wide-range of luminance.

Assume that the same f-stop will be used and only the shutter speed will be reduced to equalize the exposure.  In terms of dynamic range, would the use of a lower ISO, increase the OBSERVABLE DR at 20 by 24 inch prints from a 33 mega-pixel back? (SINAR 75LV).

I am interested also, in using the responses to see if the same presumed positive effect might be seen with my Canon 1Ds Mark II, which I just discovered is capable of utilizing a ISO 50.  Is there any reason not to expect that the presumed favorable results of utilizing a 50 ISO on the DB might be seen when ISO 50 is used on a sensor half the size?

Hi Jerry,

It depends on how the ISO control is implemented. I don't know about the Sinar, but the 1Ds2 does benefit from using the 'L' setting. It actually resembles ISO 75-80 when looking at the sensor signal level, but it does reduce noise a little and thus boost dynamic range. On my 1Ds3 there is no benefit to using "L"over 100, in fact it's the same signal level as 100, but overexposed.

Cheers,
Bart

[JerryReed]

Bart,

Thanks, I don't know how I might have found that arcane information on the differences between the ISO performances between the 1 DsM2 and Mark III, if you had not responded.  Really great stuff.  Thanks very much.  I just love this site.

Does anyone have any information on the properties of the SINAR 75?  Theirry, are you still there?

Jerry

Hi Jerry,

It depends on how the ISO control is implemented. I don't know about the Sinar, but the 1Ds2 does benefit from using the 'L' setting. It actually resembles ISO 75-80 when looking at the sensor signal level, but it does reduce noise a little and thus boost dynamic range. On my 1Ds3 there is no benefit to using "L"over 100, in fact it's the same signal level as 100, but overexposed.

Cheers,
Bart

[ErikKaffehr]

Hi,

Just a small comment. I'd suggest that ISO matters little as long as your exposure is to the right. So I guess exposing +1 stop at 100 ISO and +/- 0 at ISO 50 will give the exactly same results. On DSLRs there seems to be more manipulation on preamplifiers according to ISO so they may differ a little. Anyway, both MFDBs and DSLRs probably will give you more DR than what you can show in print, like 12 stops. A print has a DR of about 7 stops, so you need to adjust tonality (gradation curve) so you get an esthetic result.

The main issue with digital capture is the lack of "shoulder" effect in the highlights. Reducing ISO will certainly not help with that. If your subject is stationary you could try to use HDR, shooting multiple pictures  with different exposures, a technique that is supported in Photoshop and many other tools.

Best regards
Erik

What is your experience with utilizing an ISO of 50 versus ISO 100?  The application that I am considering this for is table top still life, where the subject might be lit with just one light, providing for a wide-range of luminance.

Assume that the same f-stop will be used and only the shutter speed will be reduced to equalize the exposure.  In terms of dynamic range, would the use of a lower ISO, increase the OBSERVABLE DR at 20 by 24 inch prints from a 33 mega-pixel back? (SINAR 75LV).

I am interested also, in using the responses to see if the same presumed positive effect might be seen with my Canon 1Ds Mark II, which I just discovered is capable of utilizing a ISO 50.  Is there any reason not to expect that the presumed favorable results of utilizing a 50 ISO on the DB might be seen when ISO 50 is used on a sensor half the size?

Jerry Reed

[JerryReed]

Erik,

I am unfamiliar with the term "shoulder" as it relates to the highlight portion of the curve.  Can you expand your discussion just a  bit please?

Jerry

Hi,

Just a small comment. I'd suggest that ISO matters little as long as your exposure is to the right. So I guess exposing +1 stop at 100 ISO and +/- 0 at ISO 50 will give the exactly same results. On DSLRs there seems to be more manipulation on preamplifiers according to ISO so they may differ a little. Anyway, both MFDBs and DSLRs probably will give you more DR than what you can show in print, like 12 stops. A print has a DR of about 7 stops, so you need to adjust tonality (gradation curve) so you get an esthetic result.

The main issue with digital capture is the lack of "shoulder" effect in the highlights. Reducing ISO will certainly not help with that. If your subject is stationary you could try to use HDR, shooting multiple pictures  with different exposures, a technique that is supported in Photoshop and many other tools.

Best regards
Erik

[JonathanBenoit]

Anyway, both MFDBs and DSLRs probably will give you more DR than what you can show in print, like 12 stops. A print has a DR of about 7 stops, so you need to adjust tonality (gradation curve) so you get an esthetic result.

I am assuming you meant 35mm DSLRs in the above statement.. Not to get too deep into this, but MF is about 12, while 35mm DSLRs are more like 8 stops.

[ErikKaffehr]

No,

Prints have a DR of about 7 stops, period. DSLRs have about one stop less DR than MFDBs according to photon statistics. I'm fully aware that MF is attributed to have several stops of DR than DSLRs, but I have never seen it substantiated by measurements or useful examples.

DR is mostly limited by shot noise and that is essentially dependent on the number of photons detected by the sensor. The number of photons detected by the sensor are essentially dependent on the size of the sensor. The MFDB sensors are about twice the size of the full frame sensors in DSLR. Doubling the size will about 1/2 stop advantage, to my understanding. This is just physics/simple math. MF sensors actually come in different sizes so some may be larger.

Anyway, prints have a very limited DR, lets assume DMax is 2.3 and Dmin is 0.15. So DR is 2.3 - 0.15 that is 2.15. To make it into steps you divide with log(2) which is pretty exactly 0.3, so dynamic range is about 2.15 / 0.3 -> 7 steps. Anything more than that you need to compress to fit into a print.

Now, there is an "engineering" definition of DR, and that is Maximum signal divided by signal at SNR = 1. That definition is essentially dependent on read noise which used to be lower on CMOS DSLRs using correlated double sampling than on CCDs. Using that standard benefits DSLRs and also may explain why they work better at high ISO, simply because they have less electronic noise. Using that engineering definition of DR may help DSLRs to further reduce the one stop advantage that MFDBs have based on photon statistics.

I have only seen one comparison of DR between MFDBs and DSLRs and that was in a Swedish periodical, quite a few years ago. The were shooting a Stouffer edge with a dozen different cameras and checked the darkest field that was resolved. The difference was about one step according to that measurement.

Best regards
Erik

I am assuming you meant 35mm DSLRs in the above statement.. Not to get too deep into this, but MF is about 12, while 35mm DSLRs are more like 8 stops.

[JerryReed]

Erik,

I am unfamiliar with the term "shoulder" as it relates to the highlight portion of the curve. Can you expand your discussion just a bit please?  Are you saying that the slope from say 255 to the quarter tone should be relatively flatter (less slope)?

So, would you be looking to change the gamma more toward the shadows, to avoid making the mid-tones overly contrasty?

Jerry

[EricWHiss]

I think you'll see bigger differences between the two cameras than you would with a 1 stop ISO change.  Definitely the Sinar will be best at base ISO and be hugely better than the 1D3 in terms of DR.   But if its still life, why not shoot multiple exposures and blend?

[ErikKaffehr]

OK,

A real problem with digital capture is that sensors are linear devices. If you double exposure you get twice the signal. The sensor cell can hold about 50 000 electrons and we can assume that each captured photon (particle of light) results in one electron. This is just fine.

The problem is that when we start saturating the sensor say reaching 49999 electrons the sensor can accept one more electron. If you increase exposure by a stop there would be 100000 electrons but the sensor cannot hold that, so all additional information will be lost. Negative film has a shoulder effect. If you increase exposure beyond "saturation" density will still increase but at a much slower rate than at lower exposure. This behavior is called shoulder on the gradiation curve. So when you overexpose on a digital sensor it will clip, film will saturate slowly.

To achieve maximum DR you want to expose so that non-specular highlight is within the capability of the sensor but you want to maximize the photons detected so you get minimum noise in relation to signal, therefore you want to "expose to the right". As long as you expose to the right ISO setting matters little, as long as the ADC (Analog Digital Converter) can resolve the dynamic range of the sensor signals. DSLRs tend to have adjustable preamplifiers  and they may play a role. Most MFDBs use fixed pre amps and may have larger dynamic range on ADC (Phase is supposed to have 16-bit ADC, whereas DSLRs use 12 or 14 bits).

Read noise on Kodak and DALSA CCDs tends to be in the low tens of electrons while DSLR sensors seem to be around 3 electrons or so. "Professional" sensors seem to be able to hold about about 50 000 electrons per pixel weather CCD or CMOS, DSLR or MFDB.

Sorry, this is not as exhaustive as it could be, but may give some insight in the issues involved.

You may check this article for more detail: http://www.clarkvision.com/articles/digita...mary/index.html

There is also an excellent article by "ejmartin" (Emil Martinec?) , a frequent contributor to this forum, but I cannot find a link to it right away. Update, it's here: http://theory.uchicago.edu/~ejm/pix/20d/te...oise/index.html .

In this excellent demo http://www.zacuto.com/shootout (episode 1 at 15:30 they have a scene including a candle of lighting and a lightbulb). On film you can see the fingerprints on the lightbulbs with the DSLRs the lighbulb turns solid white, check it!

Best regards
Erik

Erik,

I am unfamiliar with the term "shoulder" as it relates to the highlight portion of the curve.  Can you expand your discussion just a  bit please?

Jerry

[bjanes]

Erik,

I am unfamiliar with the term "shoulder" as it relates to the highlight portion of the curve. Can you expand your discussion just a bit please?  Are you saying that the slope from say 255 to the quarter tone should be relatively flatter (less slope)?

So, would you be looking to change the gamma more toward the shadows, to avoid making the mid-tones overly contrasty?

Jerry

Click on this link for an explanation of Knee and Shoulder of an H & D curve.

H & D Curve

[bjanes]

There is also an excellent article by "ejmartin" (Emil Martinec?) , a frequent contributor to this forum, but I cannot find a link to it right away.

Erik

Try this link for Emil's treatise:

Noise, Dynamic Range and Bit Depth

[ErikKaffehr]

Thanks! Exactly the article I was looking for!

BR
Erik

Try this link for Emil's treatise:

Noise, Dynamic Range and Bit Depth

[EricWHiss]

I forgot to mention that I can easily see the differences in DR between the MFDB and DSLR in prints made on my z3200 printer.  For some subjects like flowers or black and whites its clear that the MFDBs handle subtle color transitions better.  If you had to shoot only single shot and needed to record a high contrast scene then your Sinar back even at ISO 400 may produce a better file than your DSLR in the final print.  It's interesting because the noise kind of gets downplayed in prints but the DR does come through.

[bjanes]

Bart,

Thanks, I don't know how I might have found that arcane information on the differences between the ISO performances between the 1 DsM2 and Mark III, if you had not responded.  Really great stuff.  Thanks very much.  I just love this site.
Jerry

If you have a program that looks at raw pixel values (such as Iris or Rawnalize), you can determine how the camera handles the ISO setting. The essential thing is to keep actual exposure the same while varying the ISO setting on the camera. For example, use the same shutter speed and f/stop and take shots of a white wall at various ISOs. If the raw values are the same, then the camera is not really changing the amplification. It is instructive to take a number of exposures up to clipping. You can do this by changing the shutter speed or f/stop and keeping the lighting the same. Since shutter speeds with modern electronic shutters are more accurate than f/stop settings (which have a mechanical linkage), it is best to vary the shutter speed. A convenient way to take multiple exposures with only one exposure is to use a step wedge such as the Stouffer wedge.

Here is the Stouffer wedge that is over-exposed so that clipping occurs in steps 1 through 3 at a given exposure, 1/13 sec @ f/8 in this case.
[attachment=22450:May282008_Img_0007.jpg]

You can vary the ISO and see what happens to the pixel values. With the Nikon D3, here is what I get. The base ISO of the camera is ISO 200, but it does offer a low ISO setting of 100. The pixel values for ISO 100 and ISO 200 are virtually the same for this camera. With exposure held constant, setting the camera to ISO 100 merely causes overexposure and can lead to clipping.
[attachment=22449:ISOs.gif]