“Photographic Dynamic Range” of Kodak Vision3 50D film 5203 / 7203

[ilza]

Time to time, we are getting into discussions of the Dynamic Range of film. Of course, the result depends vastly on what is the definition of Dynamic Range! For the purpose of this post, I’m going to use the definition of “Photographic Dynamic Range” from Photons2photos. (This is just an exercise without any specific purpose. However, I hope that somebody may find this useful — or may find errors in my approach. ;-)

The definition, essentially, says: the useful range is where the SNR on a circle of diameter 22μm is above 20. (Per comments on this definition, one can recalculate to different circles; for example, with 48μm SNR must be above 43.6 — and Kodak measures density noise on such circles! Note that 1/43.6=0.0229.)

(No, I do not think this definition makes a lot of sense. And I have no idea whether it is reasonable in the analog context. However, this is the only definition I know which allows comparison of digital and film on “somewhat equal footing”. Caveat emptor.)

I digitized the data for noise and the tone-curve from Kodak’s datasheet (I will attach it later). (To estimate γ [such that 1/γ is the slope of tangent in the Kodak’s log/log plot of density-vs-exposure], I would ever look for y-intercept of the tangent, or would estimate the slope directly.) This allows recalculation of the Kodak’ σ(density) to Photons2Photos SNR, or σ(log(photon count)).

Here are the results: the useful exposure is (where noise is below 0.0229):

1.7–4.27 (base 10), or 2.57*log₂10=8.54 stops.

In fact, one may want to recalculate it for different film formfactors. Below, we compare digital-FF vs film-IN-THE-GIVEN-FORMFACTOR.

IMAX [15 sprockets] (rescaling 2x):

Allow noise 0.0458. The range is 1.17–4.61, or 3.44*log₂10=11.42 stops.

6×9cm (rescaling 2.5x):

Allow noise 0.0573. The range is 1.0–4.75, or 3.75*log₂10=12.46 stops.

4×5in (rescaling 4x):

Allow noise 0.0916. The range is 0.94–5.16, or 4.22*log₂10=14.02 stops.

8×10in (rescaling 8x):

Allow noise 0.183. The range is 0.92–5.56, or 4.64*log₂10=15.41 stops.

(The last two require extrapolation beyond Kodak’s max=5.0. If one does not extrapolate beyond 5, the range goes down to 13.49 and 13.55 correspondingly. My extrapolation does not take into account solarization — just continues the exponential growth of noise observed with exposures in the range 4.3–5.0. So this extrapolation is, in fact, very suspect.)

 

[ilza]

Here is the promised digitization (BTW, it is NEGATIVE 50ISO film!):

# exposure density y-intercept γ σ(R,G-average) noise
### (the density at 0.95 is calculated based on γ, which was measured separately)
### We keep extra meaningless digits too!
.95 0.627 0.5 7.5 0.0033 0.0684
1.03 0.64 0.4 4.29 0.0043 0.0510
1.22 0.72 0.2 2.35 0.0067 0.0435
1.81 1.04 0 1.74 0.0046 0.0221
# 1.323–2.99 0.8–1.8 - 1.676 0.0045 0.0209
# 2.99–3.9 1.8–2.3 - 1.82 0.0045–0.004 0.0209–0.0201
3.865 2.28 0.18 1.84 0.004 0.0203
4 2.36 0.28 1.923 0.0037 0.0197
4.3 2.514 1 2.84 0.0032 0.0251
4.54 2.57 1.6 4.68 0.00305 0.0394
4.81 2.6 2 8.02 0.0028 0.0620
5 2.64 2.18 10.87 0.0028 0.0841

Below 0.95 I used extrapolation (I noted that γ ≈ γ₀ + const/(D-0.9) below 1.81). Above 4.3 the noise grows exponentially (when it depends on log(D)) with exponent 5.625. (Note that for digital, the exponent would be less than 1, 1/√10.) This allows extrapolation too…

σ(R,G-average) is the average of noise in R- and G-channels. The noise in B-channel is much higher, but since human eye is not very sensitive to intensity of B-channel, I think it is prudent to mostly ignore the B-channel. Instead, I multiply the average noise of R- and G-channels by a constant slightly larger than 1 (I choose c=1.2). (Since B-noise is yet higher when exposure is below 1.3, one could use higher c in this region. However, I used c=const.)

For noise, I use formula log(10)·γ·c·σ. Ouph, looks like I gave you all the details now!

Update: forgot to say: density, y-intercept and γ are for the G-channel.

 

[ilza]

Forgot to say: when recalculating to larger formfactors, I mindlessly used the law which works for digital sensors — but most probably does not work for film: “SNR depends linearly on the diameter of the window”. For film, the growth is much slower (and eventually, SNR saturates!).

However, in absence of any hard data on this growth, the formula I used is “as good as it gets”. For IMAX and 6×9cm this is very reasonable (since this just “compensates” our rescaling from 22μm to 48μm). For 4×5in and 8×10in it should be taken with “another grain of salt” (in addition to our extrapolations!).

 

[ilza]

Forgot about one more (the most important? ;-) ) formfactor:

Super35 (rescaling 0.69x)

Allow noise 0.0158. The range is ∅, or 0.00 stops.

 

[NateW]

To be honest, I really don’t understand the math and everything so forgive me if I’m not understanding what you were trying to say, but can’t you just look at the curves on the data sheet and know that at the rated speed the dynamic range where detail will be recorded in the negative extends from about -6 to +8 for about 14 stops?

 

[ilza]

… can’t you just look at the curves on the data sheet and know that at the rated speed the dynamic range where detail will be recorded in the negative extends from about -6 to +8 for about 14 stops?

Hmm, could not you just look at my conclusion? Which says: for Super35 format (the intended use of this Kodak film) the dynamic range is 0 stops. Kinda far from your “14 stops”, is not it?!

… Anyway, the key words of your are “where details will be recorded”. Think: what does it mean?

The contemporary approach is: where the noise is not corrupting signal “too much”. So you cannot determine this by looking at one curve: you need two, one for transmission of signal, and one for the added noise. So one must use the plots on the left in the datasheet (and this is why I was forced to use decimal log, instead of “stops”).

(And this also explains why the result depends on the formfactor: if the same signal is distributed over larger area, it is harder to corrupt it by noise.)

The final nail is that to get hard data, we need to quantify what this “too much” (2 paragraphs back) actually means. What I did above was to use a particular widely used metric for estimates of dynamical range of digital cameras.

(I do not say that IMO this metric makes sense — but with more sensible metrics the disadvantages of film [in small formfactors] would be yet more pronounced.)

TL;DR: to estimate the dynamic range of digital cameras, there is a certain cut-off on the quality of produced image. “PDR” is the range of brightness where the quality is better than this cutoff. “The best film” (AFAIK) would NEVER produce such quality when used in the Super35 formfactor. So its dynamic range is 0 stops.

 

[cameronrad]

… can’t you just look at the curves on the data sheet and know that at the rated speed the dynamic range where detail will be recorded in the negative extends from about -6 to +8 for about 14 stops?

Hmm, could not you just look at my conclusion? Which says: for Super35 format (the intended use of this Kodak film) the dynamic range is 0 stops. Kinda far from your “14 stops”, is not it?!

… Anyway, the key words of your are “where details will be recorded”. Think: what does it mean?

The contemporary approach is: where the noise is not corrupting signal “too much”. So you cannot determine this by looking at one curve: you need two, one for transmission of signal, and one for the added noise. So one must use the plots on the left in the datasheet (and this is why I was forced to use decimal log, instead of “stops”).

(And this also explains why the result depends on the formfactor: if the same signal is distributed over larger area, it is harder to corrupt it by noise.)

The final nail is that to get hard data, we need to quantify what this “too much” (2 paragraphs back) actually means. What I did above was to use a particular widely used metric for estimates of dynamical range of digital cameras.

(I do not say that IMO this metric makes sense — but with more sensible metrics the disadvantages of film [in small formfactors] would be yet more pronounced.)

TL;DR: to estimate the dynamic range of digital cameras, there is a certain cut-off on the quality of produced image. “PDR” is the range of brightness where the quality is better than this cutoff. “The best film” (AFAIK) would NEVER produce such quality when used in the Super35 formfactor. So its dynamic range is 0 stops.

13 stops.

 

[ilza]

13 stops.

Wrong. The correct way to report this would be

“In their advertisements, Kodak claims 13 stops.”

Note the difference with what I wrote:

According to “this and this methodologies”, what the Kodak datasheet contains means 0 stops.

Can you see the difference?

 

[cameronrad]

According to “this and this methodologies”, what the Kodak datasheet contains means 0 stops.

Can you see the difference?

Then maybe those methodologies aren't applicable to determining the dynamic range of color film.

http://dicomp.arri.de/digital/digital_systems/DIcompanion/index.html

http://last.hit.bme.hu/download/fir...Color_Management_Encoding_Solutions__2009.pdf

http://www.shutterangle.com/2012/cinematic-look-dynamic-range/

https://www.dantestella.com/technical/dynamic.html

 

[ilza]

maybe those methodologies aren't applicable

Well, I did manage to apply them — so they must be “applicable”?…

Anyway, jokes aside, I have no clue what kind of meaning of “not being applicable” you could have used above. One thing I did show is very simple:

• Take “the minimal acceptable quality” used for estimates of dynamic range.
• Digital cameras produce this quality in the “shadows” end of DR.
• The “best” film produces a worse quality in Super35 for any exposure.

What may be “not applicable” here?!

(BTW, thanks for the references! I suspect I might have seen them all — but I will double-check.)

 

[cameronrad]

maybe those methodologies aren't applicable

Well, I did manage to apply them — so they must be “applicable”?…

Anyway, jokes aside, I have no clue what kind of meaning of “not being applicable” you could have used above. One thing I did show is very simple:

• Take “the minimal acceptable quality” used for estimates of dynamic range.
• Digital cameras produce this quality in the “shadows” end of DR.
• The “best” film produces a worse quality in Super35 for any exposure.

What may be “not applicable” here?!

(BTW, thanks for the references! I suspect I might have seen them all — but I will double-check.)

Look at the examples of films shot on super35:

https://en.wikipedia.org/wiki/Super_35#Examples

The quality is fine and the dynamic range is most definitely more than 0 stops.

 

[alanr0]

Forgot about one more (the most important? ;-) ) formfactor:

Super35 (rescaling 0.69x)

Allow noise 0.0158. The range is ∅, or 0.00 stops.

An interesting comparison, but you seem to assume that Bill Claff's 0.022 mm CoC applies to a 36 x 24 mm full frame sensor.

Bill makes clear on photonstophotos that this is for a Nikon D300 APS-C camera with a 28.4 mm sensor diagonal. Here CoC diameter is 1/1291 sensor diagonal.

For a 36 x 24 mm FF sensor (43.3 mm diagonal), the equivalent CoC is 0.0335 mm.

The acceptable SNR for a 48 μm aperture on 36x24 FF is 30.5, not 43.6.

Super 35 frame is 24.89 x 18.66 mm, with 31.1 mm diagonal, so equivalent CoC is 24 μm, 48 μm acceptable SNR is 40, with allowed noise 0.025 - slightly larger than for APS-C.

Dynamic range will be roughly 8.6 stops.