Scanner as color densitometer

[koraks]

The most of the work is in calibrating the thing

Yeah. And that's going to be one hell of a job, I can tell you that. Having worked on similar devices in a number of ways, I can assure you that your idea looks simple enough on paper, but gets really complicated once you put it in practice.

It would make more sense to wait a bit until @dkonigs introduces a color densitometer. His approach is also a lot more sensible, using an AMS I2C color sensor. I've worked with similar sensors and they are far superior to a CMOS camera module or a phototransistor or photodiode (yes, have worked with those, too; specifically a few types from Hamamatsu paired with integrating ttansimpedance amplifiers). Discrete devices work great and can perform better than integrated solutions, but the "calibration" issue (which is about much more than just that) will explode in terms of complexity. Been there, done that. I'll spare you the details.

Electronics is a great hobby, but keep in mind every hour spent with a soldering iron is an hour not spent on making images. Too bad there are only 24 hours in a day, but we'll have to put up with that, I guess. So choose wisely...

 

[dkonigs]

Yeah. And that's going to be one hell of a job, I can tell you that. Having worked on similar devices in a number of ways, I can assure you that your idea looks simple enough on paper, but gets really complicated once you put it in practice.

It would make more sense to wait a bit until @dkonigs introduces a color densitometer. His approach is also a lot more sensible, using an AMS I2C color sensor. I've worked with similar sensors and they are far superior to a CMOS camera module or a phototransistor or photodiode (yes, have worked with those, too; specifically a few types from Hamamatsu paired with integrating ttansimpedance amplifiers). Discrete devices work great and can perform better than integrated solutions, but the "calibration" issue (which is about much more than just that) will explode in terms of complexity. Been there, done that. I'll spare you the details.

Yes, calibration is absolutely the hard part. Its something I haven't really cracked yet, but I finally have most of the equipment necessary to pull it off. Without it, you can probably get "in the ballpark," but its not hard to end up in a situation where your readings totally mismatch the existing (albeit mostly discontinued) products.

You basically take all the problems of a B&W densitometer, and add another dimension.

In my research, though, I have stumbled across numerous papers where people have attempted to use RGB digital cameras of various sorts to measure reflectance spectra, so at least its a problem someone has partially thought about before.

But to do all of this calibration nonsense, you really need a spectrometer, a monochromator, a broadband light source, and all of the bits and bobs to put them together for a wide variety of measurements. To make matters worse (or simply more annoying), you really can't find "spectrally calibrated" reference materials for transmission measurements. (Though you can easily find these for reflection measurements.)

 

[Steven Lee]

@dkonigs you're working on a color transmission densitometer?

 

[dkonigs]

@dkonigs you're working on a color transmission densitometer?

I am, but I always like to cautiously state that there's absolutely no guarantee that I'll be successful.

Building the hardware is the easy part, but getting the right readings out if it is the hard part.

 

[pwadoc]

Yeah. And that's going to be one hell of a job, I can tell you that. Having worked on similar devices in a number of ways, I can assure you that your idea looks simple enough on paper, but gets really complicated once you put it in practice.

It would make more sense to wait a bit until @dkonigs introduces a color densitometer. His approach is also a lot more sensible, using an AMS I2C color sensor. I've worked with similar sensors and they are far superior to a CMOS camera module or a phototransistor or photodiode (yes, have worked with those, too; specifically a few types from Hamamatsu paired with integrating ttansimpedance amplifiers). Discrete devices work great and can perform better than integrated solutions, but the "calibration" issue (which is about much more than just that) will explode in terms of complexity. Been there, done that. I'll spare you the details.

Electronics is a great hobby, but keep in mind every hour spent with a soldering iron is an hour not spent on making images. Too bad there are only 24 hours in a day, but we'll have to put up with that, I guess. So choose wisely...

I tend to disagree that any time not spent making images is wasted time. I've built so many tools for photography, and every one of them has taught me something that I've been able to make use of, and usually I get a useful tool out of it too. Plus there's the fact that photographic equipment tends to be insanely overpriced for what it is. If I didn't build stuff myself I'd have a pretty limited set of equipment at this point, and I think this is another such case. The fully featured equipment is expensive, so maybe there's a way to fulfill this need without a ton of expense.

I've played around with one of those AMS devices, but they might actually be overkill for the purpose of a very specialized instrument for measure the status M density of film test strips. The more complex the the sensor, the more work you have to do characterize the response. It's super helpful if you're trying to build a fully-featured instrument, But if you're just trying to measure the transmission of narrow band LED with a known emission peak, you can get away with a simple sensor that just ignores color altogether. All you care about is brightness, and in the simplest case the absolute measurement doesn't even matter. As long as you get the same measurement on a given patch of a test strip relative to the reference strip.

Yes, calibration is absolutely the hard part. Its something I haven't really cracked yet, but I finally have most of the equipment necessary to pull it off. Without it, you can probably get "in the ballpark," but its not hard to end up in a situation where your readings totally mismatch the existing (albeit mostly discontinued) products.

You basically take all the problems of a B&W densitometer, and add another dimension.

In my research, though, I have stumbled across numerous papers where people have attempted to use RGB digital cameras of various sorts to measure reflectance spectra, so at least its a problem someone has partially thought about before.

But to do all of this calibration nonsense, you really need a spectrometer, a monochromator, a broadband light source, and all of the bits and bobs to put them together for a wide variety of measurements. To make matters worse (or simply more annoying), you really can't find "spectrally calibrated" reference materials for transmission measurements. (Though you can easily find these for reflection measurements.)

Sounds like an interesting project! How does your design work? Seems like you're trying to measure the transmission/reflection of a pretty broad spectrum? I know that you can still buy calibration targets for densitometers, and interestingly the one I own calibrates itself off of a grey step chart. Maybe something like an transparent IT-8 test target could be useful in that regard?

 

[koraks]

I tend to disagree that any time not spent making images is wasted time.

Hey, I didn't say it's a waste of time Like you, I regularly make photography-related trinkets and devices. It's great fun and I agree with you the out-of-pocket expense is often less attractive than spending a few hours on a problem. But keep in mind that the vast majority of the people on this forum are enthusiastic about photography, but don't always have the same combined electronics + photography hobby that you and I appear to share. Not everyone is in it for the same set of reasons - at least that's my experience. And my guesstimate is that Steven Lee isn't looking to whip out a soldering iron and an Arduino to get his color process calibrated. My guess is that he's trying to get as far as he possibly can with readily available equipment that requires the least amount of modification.

those AMS devices, but they might actually be overkill

Nah, it's the other way around. Give it a try, you'll see what I mean. Take for instance a photodiode, which is what you'd typically use to measure light intensity with good linearity and sensitivity. If you build something around a diode to do density measurements on C41 and E6 film, you'll find that in practice, it's actually very hard to get good linearity out of a photodiode setup, especially over the substantial dynamic range you'd need. Moreover, you'll be battling all manner of noise issues. Those AMS devices carry all the solutions to those problems. They're far more straightforward to use in such a case than a raw diode and some homemade amplification and conversion circuitry. Let alone a phototransistor, which suffers from the same problems as a diode, only much worse.

There's a project that I've bookshelved for now, but that I'll pick up on later (I hope) that revolves around something very similar. In this, I've tried several approaches, mainly revolving around either AMS sensors or around discrete photodiodes. I've built dozens of circuits already around the latter. Again, I'll not go into detail because I'd derail this thread with many pages of schematics and analysis. Just to give you one example: have you ever considered what the different charge injections are of a CMOS switch, a small-signal MOSFET and an RF BJT are into the input of a suitable opamp in an integrating transimpedance amplifier, and how that influences the linearity of the analog signal across a useful light intensity range? These are the kinds of effects you'll end up juggling if you go down the path that you think is so simple. I thought that too, at some point Trust me, if you want to try something like this, you'll learn to appreciate integrated solutions like the AMS line of sensors. They make life so simple - which you only realize once you know how complex it really is. The many circuits I've designed and tested involving diodes over the timeframe of several months never even came close to the kind of linearity that the AMS solution provided in a breadboard setup that literally takes 5 minutes to set up.

Anyway, enough said on this; let's allow the thread to go back to Steven's question and the more realistic (IMO) solutions. We'll get back to this at some point, and it'll be fun - stick around

Without it, you can probably get "in the ballpark," but its not hard to end up in a situation where your readings totally mismatch the existing (albeit mostly discontinued) products.

In know what you mean. You've not made things easy on yourself by including the requirement of matching the behavior of existing products, although it's probably difficult to avoid this for a commercial product. Having said that, for a situation like this one, a device with linear and consistent output and sufficient sensitivity over a relevant dynamic range would already be sufficient, even if it doesn't track with existing equipment in the field. As long as the measurements are repeatable on the same machine, it would already work. Have you considered launching a kind of beta-version of your unit that does this? I can imagine that this would already be useful to many, and you could always follow up with software update once you fix the puzzle, and perhaps a conversion app/worksheet that translates the measurements you can make now to future ones that track well with old equipment.

 

[nmp]

Educate me....isn't color densitometer the same as regular densitometer with filters?

:Niranjan.

 

[nmp]

Speaking of Faust target, I tried to calibrate my scanner with one of those and I decidedly got green tinge from the resulting profile. So the target does not work well - at least with Vuescan. I do have X-Rite Colormunki Photo (or i1Studio as it is called now) that can also calibrate a scanner with their ColorChecker (which should be more accurate than Faust) target. My scanner as is seems close enough so I haven't bothered anymore. May be one of these days I will check it out.

:Niranjan.

 

[koraks]

I tried to calibrate my scanner with one of those and I decidedly got green tinge from the resulting profile.

Which software app did you use? I used the old iphotoicc app. It worked, but I also have some doubts about the quality of the profile, although it seems slightly better than the built-in Epson profile. Btw, this was with a reflective target.

 

[nmp]

Which software app did you use? I used the old iphotoicc app. It worked, but I also have some doubts about the quality of the profile, although it seems slightly better than the built-in Epson profile. Btw, this was with a reflective target.

Vuescan - it has a scanner profiling option. I also used the reflective targets - IT8.7/2 both on Kodak and Fuji Crystal. I am not familiar with iphotoicc.

By the way, the way I see it (and I could be wrong) but the "luminosity" value that the EpsonScan calculates is not the true "luminance" that is used to calculate density. The former is how our eyes perceive color (which is a weighted average of the three components according to sensitivity) whereas the latter is the more objective true representation of the light being reflected. Some of the error may have been due to that. I don't know how Vuescan does it.

The way I understand, to get correct density one would have to convert the RGB value to Lab using something like this with the correct colorspace (go to CIE Calculator):

Welcome to Bruce Lindbloom's Web Site

www.brucelindbloom.com

Then convert the L value to D using something like this:

Sign in - Google Accounts

:Niranjan.

 

[Rolleiflexible]

Search for a used xRite i1Pro. Yes, they cost a pretty penny, but mine I got for about half the price you mentioned and it came with everything

A far easier and cheaper solution, for linearizing alt process negatives, is a $60 gadget called a Color Muse, sold in hardware stores to match paint colors. It links by Bluetooth to a phone app and gives LAB values to step chart wedges. Credit Clay Harmon for this hack -- he's tested it against i1 Pros and found it accurate for these uses. Here's a link to Clay's work:

Color Muse: an inexpensive colorimeter to use for digital negatives

clayharmonblog.com

Clay has even posted a step chart resized to use with a Color Muse, and a web app for collecting and tabulating the Color Muse output. Having used both the i1 and the Color Muse, I will never go back to the i1.

 

[koraks]

A far easier and cheaper solution, for linearizing alt process negatives, is a $60 gadget called a Color Muse

Yes, I read about it some weeks after I got my i1Pro
I haven't looked into availability of the Muse in Europe.

 

[dkonigs]

Educate me....isn't color densitometer the same as regular densitometer with filters?

:Niranjan.

It can be, provided that you have a broadband light source that covers the full spectrum, a sensor that's sensitive to the full spectrum, and filters that have the specific response curves mentioned in the standards that describe how densitometers are supposed to work (e.g. ISO 5-3:2009).

Some color densitometers actually work this way, by having a filter wheel you can rotate through a variety of options depending on which color range you want to measure.

Other color densitometers have a ring of sensors with each one being fitted with a different custom filter so they can read everything at once.

The AMS sensors mentioned above are kinda like a microscopic version of the second approach. A grid of sensors each of which has a filter tuned to a specific wavelength range, all integrated into one tiny package. The problem is that these wavelength ranges don't actually match the ones specified in the ISO spec for densitometers. Thus, some of the complicated calibration problems where you need to characterize them and figure out how to extrapolate their data into the data you want.

 

[nmp]

It can be, provided that you have a broadband light source that covers the full spectrum, a sensor that's sensitive to the full spectrum, and filters that have the specific response curves mentioned in the standards that describe how densitometers are supposed to work (e.g. ISO 5-3:2009).

Some color densitometers actually work this way, by having a filter wheel you can rotate through a variety of options depending on which color range you want to measure.

Other color densitometers have a ring of sensors with each one being fitted with a different custom filter so they can read everything at once.

The AMS sensors mentioned above are kinda like a microscopic version of the second approach. A grid of sensors each of which has a filter tuned to a specific wavelength range, all integrated into one tiny package. The problem is that these wavelength ranges don't actually match the ones specified in the ISO spec for densitometers. Thus, some of the complicated calibration problems where you need to characterize them and figure out how to extrapolate their data into the data you want.

Thanks @dkonigs for giving a little "color" on my query. As usual the devil is in the details.

On a related subject, if you have the L*ab coordinates of a patch - can you not get the R,G,B densities? Theoretically, the set should contain how much of each component is in the combined spectrum. In Photoshop, for example, I can take a single L*ab patch, split into the 3 individual channels, measure the individual L's for each channel and then calculate D based on those.

:Niranjan.

 

[koraks]

if you have the L*ab coordinates of a patch - can you not get the R,G,B densities?

Yes.

Theoretically, the set should contain how much of each component is in the combined spectrum.

But the real world, including chromes and negatives, does not consist of tristimulus color generators The electromagnetic spectrum is continuous. That's the tricky bit, here.

Conversions are possible in some cases, and result in compromises in some of those cases. If a conversion is meaningful depends on the context and application.

 

[dkonigs]

One thing that's important to keep in mind, is that with a proper color densitometer, "Red" isn't just red.
"Red" is actually a sensitivity curve that's centered around 620nm for "Status A" density, and centered around 640nm for "Status M" density. (Similar thing for the other colors.) So with a naive RGB sensor or conversion, you can absolutely get RGB densities, but you absolutely will not get RGB densities that actually match a benchtop densitometer on color film or paper.

(FYI, "Status A" is typically used for measuring positives, while "Status M" is typically used for measuring negatives.)

Its easy to be a bit lax on this when you're measuring B&W materials, because their response is somewhat flat across the spectrum. (Not completely flat, but close enough that it often doesn't matter.) But with color materials, including color materials that appear to contain a strip of grey patches, there's a lot more spectral variation in how light transmits or reflects.

 

[Steven Lee]

Looks like the solutions for color densitometry lay on a spectrum, with the hardware doing all the work on one end, and clever processing of signals from imperfect sensors on another.

Couple of years ago I was in fully analog beast mode, ready to dump all my digital equipment. Instead of selling my digital cameras I briefly toyed with an idea of sending them to one of those services that strip them of color array to create a high resolution + high speed B&W capture device. I now regret not doing this, because it would allow experimenting with color filtration. Could be useful for "scanning" as well.

 

[Alan Edward Klein]

How do you use densitometer and colorimeter readings to improve photography?

 

[Bill Burk]

How do you use densitometer and colorimeter readings to improve photography?

You don't think about it when you're out taking pictures (unless you are doing careful exposure metering and noting the readings).

You do this stuff before you develop the film from that day.

And you can use them while printing, as an aid.

I very rarely use these tools while printing, I rely on test strips.

But today I think I am going to use the Beseler PM2M so I can print an entire roll of film from 1980.

I want to print the whole roll and don't want to make a test strip for each print.

 

[Alan Edward Klein]

You don't think about it when you're out taking pictures (unless you are doing careful exposure metering and noting the readings).

You do this stuff before you develop the film from that day.

And you can use them while printing, as an aid.

I very rarely use these tools while printing, I rely on test strips.

But today I think I am going to use the Beseler PM2M so I can print an entire roll of film from 1980.

I want to print the whole roll and don't want to make a test strip for each print.

So if I understand what you're saying, Bill, you use the densitometer readings of a negative to determine how much light exposure time the print should get instead of using a test strip where different samples of time are used?

What about a colorimeter? How does that help?

 

[pwadoc]

Nah, it's the other way around. Give it a try, you'll see what I mean. Take for instance a photodiode, which is what you'd typically use to measure light intensity with good linearity and sensitivity. If you build something around a diode to do density measurements on C41 and E6 film, you'll find that in practice, it's actually very hard to get good linearity out of a photodiode setup, especially over the substantial dynamic range you'd need. Moreover, you'll be battling all manner of noise issues. Those AMS devices carry all the solutions to those problems. They're far more straightforward to use in such a case than a raw diode and some homemade amplification and conversion circuitry. Let alone a phototransistor, which suffers from the same problems as a diode, only much worse.

There's a project that I've bookshelved for now, but that I'll pick up on later (I hope) that revolves around something very similar. In this, I've tried several approaches, mainly revolving around either AMS sensors or around discrete photodiodes. I've built dozens of circuits already around the latter. Again, I'll not go into detail because I'd derail this thread with many pages of schematics and analysis. Just to give you one example: have you ever considered what the different charge injections are of a CMOS switch, a small-signal MOSFET and an RF BJT are into the input of a suitable opamp in an integrating transimpedance amplifier, and how that influences the linearity of the analog signal across a useful light intensity range? These are the kinds of effects you'll end up juggling if you go down the path that you think is so simple. I thought that too, at some point Trust me, if you want to try something like this, you'll learn to appreciate integrated solutions like the AMS line of sensors. They make life so simple - which you only realize once you know how complex it really is. The many circuits I've designed and tested involving diodes over the timeframe of several months never even came close to the kind of linearity that the AMS solution provided in a breadboard setup that literally takes 5 minutes to set up.

So might be missing something, but my sense is the linearity of the sensor is only going to matter if we care about the absolute value of the reading. Like, I need to take a reading with this device, and have it match up to an absolute value of density from 0-4 that I'd get on a traditional densitometer. That sounds pretty hard, and you would definitely want something with very good linearity.

What I'm suggesting is, you have your reference test strip, and you develop a test strip for comparison. You have your simple LED light with 3 emitters that emit at the status M wavelengths. You measure the test patches on the reference strip by emitting each wavelength separately through each patch, and then do the same with the comparison strip. Since you only care about relative values, you adjust the intensity of the light (you'll need a way to adjust the brightness of 3 LEDs, preferable with PWM so the peak emission frequency doesn't change too much) to ensure you're in the range with the greatest sensor linearity for each color. Then you compare those relative values. You should be able to determine with a fair degree o accuracy how much the transmission of the patches differs between the two strips, which is all we care about from a process control standpoint. If the reading on the strips are so different that the linearity of the sensor comes into play, your process is so out of control it's pointless to even run this test.

 

[dkonigs]

So might be missing something, but my sense is the linearity of the sensor is only going to matter if we care about the absolute value of the reading. Like, I need to take a reading with this device, and have it match up to an absolute value of density from 0-4 that I'd get on a traditional densitometer. That sounds pretty hard, and you would definitely want something with very good linearity.

What I'm suggesting is, you have your reference test strip, and you develop a test strip for comparison. You have your simple LED light with 3 emitters that emit at the status M wavelengths. You measure the test patches on the reference strip by emitting each wavelength separately through each patch, and then do the same with the comparison strip. Since you only care about relative values, you adjust the intensity of the light (you'll need a way to adjust the brightness of 3 LEDs, preferable with PWM so the peak emission frequency doesn't change too much) to ensure you're in the range with the greatest sensor linearity for each color. Then you compare those relative values. You should be able to determine with a fair degree o accuracy how much the transmission of the patches differs between the two strips, which is all we care about from a process control standpoint. If the reading on the strips are so different that the linearity of the sensor comes into play, your process is so out of control it's pointless to even run this test.

So if you're only measuring low/mid-density values, then taking this sort of relative comparison approach would probably work. The place these sorts of issues really come into play, however, is once you get to the higher density values. Because density is measured on a logarithmic scale, a small error can lead to a large difference in reading at the higher density values. Also, when measuring those higher density materials, you're coming right up against the limits of range on your light source and sensor.

Just to put some numbers to this, a piece of film with a transmission density of 2.5 is blocking 99.68% of the light you're trying to shine through it. A density of 4.0 means blocking 99.99% of light.

 

[pwadoc]

So if you're only measuring low/mid-density values, then taking this sort of relative comparison approach would probably work. The place these sorts of issues really come into play, however, is once you get to the higher density values. Because density is measured on a logarithmic scale, a small error can lead to a large difference in reading at the higher density values. Also, when measuring those higher density materials, you're coming right up against the limits of range on your light source and sensor.

Just to put some numbers to this, a piece of film with a transmission density of 2.5 is blocking 99.68% of the light you're trying to shine through it. A density of 4.0 means blocking 99.99% of light.

Yeah, you would probably want some very bright LEDs for this purpose, though from my experience developing an RGB light that I was using for some tests in scanning film, the light was bright enough that a fully exposed frame of slide film would still transmit a fair bit of light. Like I was testing the scanning setup with a Sekonic C-700 and I was still getting peaks of a little over 2mW/m² (per nm^-1) of intensity. For C-41 even the darkest areas shouldn't really be a challenge. It might end being more of a pain to keep the sensor linear in the low DMAX areas, or avoid blowing it out entirely, without reducing the PWM brightness so far that flicker becomes a problem.

Might try to build this thing later this week to see if it works since I actually have all the parts on hand. It would be cool to have tool that could allow photographers to compare test strips that they could build themselves for cheap.

[edit] Just tested it, at full power I can put out enough light to penetrate a fully black slide and measure the output in the linear range of the sensor I have, so this should be workable.

 

[Patrick Robert James]

I didn't see specifically what you want to do but I didn't read all the above. Life is too short. Read transmission or reflectance values?

For reflectance values just get an I1Pro. I use older software with mine. ProfileMaker Pro 5 I think. It is loaded on an old Mac. That will read in LAB and other color modes. Easy peasy. One of those doohitchamagiggies that Rolleiflexible linked to above would probably work too, but I've never used one so I couldn't say. The I1Pros aren't that expensive these days but the current software won't run with one. Xrite stopped supporting them a few years ago.

For transmission values you can use a scanner but you'll need something that is constant and calibrated to compare it to. Basically you need a test strip that is accurate and in spec, then scan that along with your new test strip at the same time. You'll be able to get the difference between the two in Photoshop or whatever program you use. There is your budget no densitometer option. Should work fine as long as your scanner is good enough.

FWIW I've been dealing with color management since the late 90s. People tend to overcomplicate it.

 

[Bill Burk]

So if I understand what you're saying, Bill, you use the densitometer readings of a negative to determine how much light exposure time the print should get instead of using a test strip where different samples of time are used?

What about a colorimeter? How does that help?

Yes, it’s not as useful as some of the better ones like RH Designs. But it can serve to help determine basic exposure

I’m only doing black and white but for color you would have a standard color negative or slide and you would make a good print from that. Then with your unknown negative, you would look for a certain tone to match (e.g., fleshtone) and read out how much difference there is between standard and the one you’re going to print.

I always do black and white so defer to “anyone else” when it comes to color.

Last night I made about 10 prints but the plan didn’t work well. I made one good print and then adjusted color channels on a module for the PM2L for three different purposes. I can’t use cyan because my Zone VI cold light is primarily cyan. Literally get no reading on the cyan channel. At first I thought the analyzer was broken.

So I set yellow to a measurement of “black” and magenta to a measurement of “white” and the white channel is zeroed so I can use it for “density”.

I got the print base time to 32 seconds and checked the different negatives. I got readings that were all over the place and basically kept the time at 32 for most of the prints. Some readings suggested 40 seconds by one channel and 20 seconds by the other, so I decided to keep 32 and they were fine.

I had one negative that was extremely overexposed (bulletproof) and I normally skipped it but this time I wanted to see what I could get. The analyzer gave me a wrong reading so the first print came out too light and I had to reprint it.

I did not think this through. I didn’t remember whether I meant black and white as in shadow and highlight… Or as in thin and dense. In my confusion I read the wrong channel.