Adventures in film characteristic analysis

[Nodda Duma]

Very cool project.

Bear in mind at some point you’ll run into accuracy limitations due to using LEDs with limited spectral content vs a full spectrum uniform source such as a Xenon Arc lamp or (more cheaply) Xenon bulb. Even white LEDs don’t cover the full spectral response of film. Small Xenon bulbs are very affordable, if cost is a concern.

So if you see unexplainable differences between your curve and those found in datasheets or between prediction and real world use, then your light sources for the test might be root cause. This will be particularly true for non-panchromatic film.

On the other hand, if accurately predicting real world performance is not required then this is a great approach. Very nice setup!

 

[radiant]

Very cool project.

Thank you so much.

I'm super aware that until proven otherwise, this is just something I've hacked together. And after that I'm going to use this only comparing between my own tests.

Whatever the result is, this is fun project to do.

 

[Nodda Duma]

Yes. Internet communication is severely flawed in that it does not accurately convey inflection. So just to be clear, I love your project and have been following with interest as I have to characterize dry plates with each new emulsion batch and sensitometry/densitometry is the way to do it. I think it’s very cool you’ve found an easy to get most of the way there. If in the future you need more accuracy, you have a path forward.

 

[radiant]

Ok, new scans from "correctly exposed" tests. I wanted to share these with you.

So what you are seeing here are two strips, exposed separately but developed + scanned together. I've shifted the lower one for one slot so you can compare the densities. Yes, you guessed right. The upper one has been exposed for one stop more than the lower. When using RGB picker and comparing the tones, they are really same. Well, you can probably see it by your eyes. Remember that these are scanned at once, so their tones are relative.

I can see 4 stops under 18%, the fift stop is a bit faint. That should be in par what HP5 can do. The upper strip has 5 stops under because of the overexposure.

For highlights or stops over 18% I can count 8 stops with different tones when using RGB picker. My scanner clearly had difficult times to scan the last slots as those are very very dense.

So what you guys think of this?

 

[Bill Burk]

I like it. Note that time-based sensitometers are considered less-than-ideal.

But I think you can do something that a typical intensity based sensitometer would have difficulty achieving. You can easily create a single strip with more than 100000:1 range. This can make your system useful for (true) solarization studies.

 

[Nikola Dulgiarov]

Radiant, very, very cool!
I think that as long as you're within reciprocity failure limits, a time-based sensitometer should be fairly accurate with most conventional emulsions, at least accurate enough for the intended purpose.

To test the intensity variation of the individual LEDs, you could set them all to the same exposure period and then check the final density for variations. Perhaps best done in triplicate, but you could get a calibration point for each individual LED (or verify that in your setup, the variation can be neglected.)

Slightly OT:
There's been a lot of discussion on using a scanner as a denitometer, and my impression is that you can either a) make rough and fast measurements using a known density reference and interpolating, or b) go into a bit more math and analysis and get a reasonably accurate absolute reading. That said, it gets me wondering about the light influx end efflux geometry. Diffuse transmission density is defined based on an ideal (lambertian) light source and a controlled cone of collection (10 degrees half-angle), or vice versa, projected light hitting a diffuse light collector/sensor. Transmission projection density (such as one encounters in a condenser enlarger system) puts tight angular limits on both the incident and trasmitted light rays, and the density result deviates from diffuse density reading as some function of the scatering of the silver in the emulsion, i.e. with respect to density (Callier effect). Would a scanner represent diffuse or projection transmission density more closely?

Slightly more OT:
I've been working on an analog densitometer project that is entirely discrete (op-amps, log transitors, etc), but suppliers still have dedicated log amplifier ICs available. A suitable photodiode (say, a blue-corrected BPW21) and a log-amp IC (say, a LOG101) could be used as the front-end to a MCU-based densitometer, taking care of the analog side of things with minimal design. A constant-current driven LED can function as a directional light source and the whole setup housed in a 3D printed enclosure that ensures the correct measurement geometry. Log-amps can be incredibly linear over 5-7 decades, and so are photodiodes in photovoltaic mode. In terms of parts cost, it is a strong motivation, if one is willing to put in design time.

 

[radiant]

I like it. Note that time-based sensitometers are considered less-than-ideal.

Once the device is calibrated, we can close our eyes from the fact that it is working time based Because results are the same, if calibrated. Which will never happen..

Here is the almost correctly exposed strip (lower one from previous). It has small flaw for base exposure (instead of 125ms, the 18% gray is exposed 100ms). But close enough for having fun analysis.

And here is graph drawn from the data:

BTW: as I know the densities of stouffer wedge, I could map the grayscale values to densities via stouffer grayscale values, I think.

 

[radiant]

Radiant, very, very cool!
I think that as long as you're within reciprocity failure limits, a time-based sensitometer should be fairly accurate with most conventional emulsions, at least accurate enough for the intended purpose.

Thank you Nikola! I'm glad you guys like this.

For the offtopics; I assume there is such things you mentioned in play when using scanner as densitometer. But what is the scale of these? I assume I can do rough hobbyist-grade analysis with scanner. I would of course never suggest that it could be used professionally. Scanned together with stouffer wedge I'm sure it is good enough for my purposes.

For doing this in analog way, that is some electronic design skills.

 

[radiant]

Of course the scanning is imporant to do right. In previous example I adjusted the film highlight end point as maximum and that caused the stouffer scale scan to be compressed. Also it isn't completely straight, here is a plot of data from previous stouffer scan:

 

[bernard_L]

BTW: as I know the densities of stouffer wedge, I could map the grayscale values to densities via stouffer grayscale values, I think.

YES! much better! addresses my previous criticism.

Plus, you get numbers out of your scanner. They are expressed in some color space, and their relation to the luminous flux passing through the film is (a) not logarithmic; (b) non-linear and complex, with many different forms floating around. See e.g. : http://www.brucelindbloom.com/index.html?Math.html; I give below a sample at random:

Make sure, however, you scan the Stouffer steps with every film you intend to measure. DO NOT trust the scanners's translation of film densities to 0-255 values to be fixed and unique.

 

[radiant]

YES! much better! addresses my previous criticism.

Make sure, however, you scan the Stouffer steps with every film you intend to measure. DO NOT trust the scanners's translation of film densities to 0-255 values to be fixed and unique.

Of course I will do that. The scanner + software can mess up my analysis big time without a stouffer. It is very valuable tool in many places, glad I bought one (well, two..).

 

[bernard_L]

Radiant, very, very cool!
Slightly more OT:
I've been working on an analog densitometer project that is entirely discrete (op-amps, log transitors, etc), but suppliers still have dedicated log amplifier ICs available. A suitable photodiode (say, a blue-corrected BPW21) and a log-amp IC (say, a LOG101) could be used as the front-end to a MCU-based densitometer, taking care of the analog side of things with minimal design. A constant-current driven LED can function as a directional light source and the whole setup housed in a 3D printed enclosure that ensures the correct measurement geometry. Log-amps can be incredibly linear over 5-7 decades, and so are photodiodes in photovoltaic mode. In terms of parts cost, it is a strong motivation, if one is willing to put in design time.

With a log transistor you still need to take care of the output being proportional to absolute temperature. If you use a dedicated log IC, you benefit from the design work of smart engineers who have taken care of this and probably other issues. Do you really need a microcontroller?
Possibly the design of the optical path and the mechanical design (pivoting arm) will take more of you time than the electronic part...
Good luck and keep us posted. Disclosure: a LONG time ago I built an enlarger-baseboard meter with a Selenium sensor, transistor log amp, and analog meter. 2.5D range, switchable offset by 1.0D.

 

[Bill Burk]

With an easel meter you can put the sample on the meter to get a better reading.

I have not been pleased with scanner as densitometer but think the best bet is scanning a Stouffer scale with the sample and confirming reading differences. VueScan does have a densitometer mode but I do not trust it over the full range. (Sorry Ed Hamrick). But it might work OK for differences between a known step and the closest sample.

I don’t mean to criticize the project for being time-based. I am planning to make a time based sensitometer using an antique Hurter Driffield disk… because I have it. Every sensitometer design misses the ideal for one reason or another.

 

[radiant]

With an easel meter you can put the sample on the meter to get a better reading.

Why didn't I think of that. I just need to check that my "slots" are wide enough for my easel meter. I build myself a densitometer but I didn't use it back then and took the sensor apart. I could calibrate that with stouffer and get real densities. But I have some software ideas ..

 

[Nikola Dulgiarov]

With a log transistor you still need to take care of the output being proportional to absolute temperature. If you use a dedicated log IC, you benefit from the design work of smart engineers who have taken care of this and probably other issues. Do you really need a microcontroller?
Possibly the design of the optical path and the mechanical design (pivoting arm) will take more of you time than the electronic part...
Good luck and keep us posted. Disclosure: a LONG time ago I built an enlarger-baseboard meter with a Selenium sensor, transistor log amp, and analog meter. 2.5D range, switchable offset by 1.0D.

My own project (diffuse density, auto zero, 6 decades) is all analog except the digital panel voltmeter which reads the density output (+1.000 V = OD 1.000). The PTAT dependence is handled by a +3300 ppm resistor in the feedback path of the logging amp. I trust that TI did a better job than me and in a much smaller space, but I enjoyed the design work ). I like analog and chose that approach; people into embedded devices, arduinos, etc, could benefit from a little analog magic at the input, rather than trying to do the compression in software. The densitometer is nearly finished and I'll be posting the schematics and design files when I get it going..

@radiant , I think your relative approach using the scanner and the step wedge is perfectly adequate. It's in a way not too different from the first visual densitometers, where one compared the relative brightness of two beams of light, one passing through the sample, and one through a calibrated variable density )

 

[radiant]

I think I have found some limits in measurement. I tried to google what is the performance of Epson V600 in stops? Epson says it is 3.4 - that is about 11 stops. Yes, that might be true but there is no mention of how linear it is. I think my previous graph hints about this. The linear part is probaly 7 stops?

Lets assume I got 12 stops (3.6) into my film density. There is no way my scanner can eat that. Or maybe I got even more..

I also noticed that my easel meter leaves the game when measuring stouffer. I get nice half stop increments on the steps 1-11 but then something happens. Probably I run out of sensor sensitivity. I think step 11 was 0.05 lux or something on my easel? And we are half way to 21. Now I realize how dense the last steps are on Stouffer ..

Well who says I need to try doing 16 stop test strips Maybe it is more reasonable to do 8 stop strips becaue hardly ever my SBR is higher than that..

 

[Nodda Duma]

Typically for characterization of imaging sensors at work, I assume the linear response to be within the region of 20% to 80% of the full dynamic range of the sensor. Outside of that I see non-linearity due to the varying sensitivity of individual pixels on the high end and shot / Johnson noise on the low end. You might be seeing this, but regardless the curves are looking decent.

 

[albada]

... The apparatus is 3D printed and has two diffusers to get even illumination to each of these strip-slots....

In your scans, illumination within each patch looks uniform. It is difficult to uniformly diffuse light, but you appear to have done it well.
Where are the two diffusers located?
And an important question: How did you prevent light under a slit from crossing through the diffuser into the neighboring slits? The only way I can think of to do this is to make a separate tiny diffuser for each slot.

Mark Overton

 

[Steve Goldstein]

Nikola, where did you get the +3300ppm/C resistor? Is anyone still making them? I haven't seen a discrete resistor like this in ages, I wish I'd kept some of the old 1k Tel Labs resistors from my first job 40+ years ago.

An analog IC would use the interconnect metal - the standard aluminum alloy is quite close to +3300ppm - and one could get a specialized sputtering target to get an even more accurate result. It's not a cheap solution, but amortized over hundreds of thousands of ICs the additional cost is negligible.

 

[Bill Burk]

I think I have found some limits in measurement. I tried to google what is the performance of Epson V600 in stops? Epson says it is 3.4 - that is about 11 stops. Yes, that might be true but there is no mention of how linear it is. I think my previous graph hints about this. The linear part is probaly 7 stops?

Lets assume I got 12 stops (3.6) into my film density. There is no way my scanner can eat that. Or maybe I got even more..

I also noticed that my easel meter leaves the game when measuring stouffer. I get nice half stop increments on the steps 1-11 but then something happens. Probably I run out of sensor sensitivity. I think step 11 was 0.05 lux or something on my easel? And we are half way to 21. Now I realize how dense the last steps are on Stouffer ..

Well who says I need to try doing 16 stop test strips Maybe it is more reasonable to do 8 stop strips becaue hardly ever my SBR is higher than that..

When you expose 3.6 on the film it will develop to about half, 1.8 Maybe you will get 2.3 but getting 4.0 on film is only a graphic arts thing.

 

[radiant]

Typically for characterization of imaging sensors at work, I assume the linear response to be within the region of 20% to 80% of the full dynamic range of the sensor. Outside of that I see non-linearity due to the varying sensitivity of individual pixels on the high end and shot / Johnson noise on the low end. You might be seeing this, but regardless the curves are looking decent.

If everything is correct otherwise, I think I can see toe and shoulder on same strip.

In your scans, illumination within each patch looks uniform. It is difficult to uniformly diffuse light, but you appear to have done it well.
Where are the two diffusers located?
And an important question: How did you prevent light under a slit from crossing through the diffuser into the neighboring slits? The only way I can think of to do this is to make a separate tiny diffuser for each slot.

Very good question. I spend some time tinkering with this. The light looks in real life really uniform and it is really coming directly upwards from the slot. If you cannot see the diffuser, then you cannot see the light. When testing it, I need to look directly above to see what is going on.

There are two diffusers: one directly above led (in its own chamber) and one halfway. And yes, the diffusers are separate. When testing first time I got the same effect you are defining here; the light leaked from slot to other if the single diffuser is used.

Also as it consist of two parts, I had some trouble with flatness of contact surfaces and I needed to print an adapter to connect the slots together without leaking light.

When you expose 3.6 on the film it will develop to about half, 1.8 Maybe you will get 2.3 but getting 4.0 on film is only a graphic arts thing.

That is true. I think I got 2.0 out of Foma 100 when developing for 2x times more.

 

[Europan]

I did this in 2007, the product I offer is called Memochrome. There are three versions of it, one with a time control (pulse duration), one with a space differentiation and one with different numbers of LED flashing at the same time. Memochrome is intended to be installed in motion-picture film printing equipment. You have 50 steps for each light colour, red-green-blue, the data are stored as CSV in Excel compatible files and read triggered by a switch built into the printer. Printing speed can be varied between 1 and 25 frames per second.

Exposure by varying time yields a graph that contains the Schwarzschild effect behaviour of a given film stock. Ilford HP 5 is made for photography also in flashlight, meaning it is corrected for short exposure times. Other films will respond quite differently.

 

[bernard_L]

With an easel meter you can put the sample on the meter to get a better reading

It's not a better reading, it's a different reading. Depends upon your intent/need.
With the sample in the negative stage, the easel reading incorporates the Callier effect for your particular enlarger illumination (diffuse/semi-diffuse, collimated). Just what you want to time an exposure or choose a paper grade. But not a standardized measurement that you can exchange over the internet.
With the sample on the meter, the input beam (from the enarger lens) is collimated and the reading is diffuse. Independent from your enlarger type. However, the opposite (diffuse/collimated) seems to be more common as a standardized measurement.

 

[radiant]

I did this in 2007, the product I offer is called Memochrome. There are three versions of it, one with a time control (pulse duration), one with a space differentiation and one with different numbers of LED flashing at the same time. Memochrome is intended to be installed in motion-picture film printing equipment. You have 50 steps for each light colour, red-green-blue, the data are stored as CSV in Excel compatible files and read triggered by a switch built into the printer. Printing speed can be varied between 1 and 25 frames per second.

Pretty cool! It must be used to automatically calibrate the process?

Exposure by varying time yields a graph that contains the Schwarzschild effect behaviour of a given film stock. Ilford HP 5 is made for photography also in flashlight, meaning it is corrected for short exposure times. Other films will respond quite differently.

An excellent and imporant note on time varying densitometers.

 

[Europan]

The deviations in light output are hardly measurable, so small that you can use the system as a sensitometer. You can make step wedges for comparisons.