VueScan as densitometer: how to calculate the CI?

[nmp]

VueScan as densitometer : Don't even think of it.

This thread prompted me to perform experiments that confirmed my suspicions. Test object: a 35mm b/w image of the Kodak Gray Scale in the Professional Photoguide. Just to avoid spurious arguments: I'm aware that the reflection densities are tricky because of specular reflections; the gray scale patch numbers are for reference only. What matters is that the same piece of film with various density patches was measured in various ways. With two different scanners, in Preview and after Scan. Then with a Macbeth TR1224 densitometer.

GSP	V700-Pre	V700-Sc	LS-2000-Pre	LS-2000-Sc	TR-1224
0.1	1.82	0.98	2.62	1.80	1.71
0.4	1.72	0.84	2.64	1.62	1.63
0.7	1.58	0.66	2.36	1.36	1.47
1.0	1.46	0.56	2.20	1.22	1.37
1.3	1.30	0.38	1.96	1.10	1.23
1.6	1.10	0.22	1.74	0.76	1.08
1.9	1.02	0.12	1.68	0.72	1.02
2.2	1.08	0.20	1.72	0.80	1.05

Legend: GSP: GrayScale Patch number on the Kodak card, again for reference only. V700-Pre: measurement after preview. V-700-Sc: after scan. Ditto for Nikon LS-2000.

The anomalous readings (higher density) for the "2.2" patch are not too surprising: being in a corner of the gray card, it may have been affected by specular reflection. When I use this card in other measurements, I discard the two darkest pactches for that same reason, and I use my measured values for the reflection densities of the other patches. The true (TR1224) densities are fairly high just because I happened to choose a frame exposed at +3EV; they are still well within the range where proper density measurements should be expected.

The vuescan density readings were jumping around by at least +/-0.05 depending on the probe position; I did not bother to attempt some kind of averaging because (a) probe size and averaging should be an option in a well designed tool; (b) the discrepancies between columns are an order of magnitude larger.

The TR1224 was calibrated using a Stouffer T3110 (calibrated) wedge. Anyway, the re-calibration from the previous Eprom-stored values involved changes of 0.01D.

My conclusion: vuescan is worthless as a densitometer. And don't get me wrong: I use vuescan for all my scans.

Hi, Bernard....very interesting.

Just so I understand correctly - how did you calculate D from the scanner data?

Also, why do you think there is such a discrepancy between preview and scan numbers - what are your resolutions for both. Preview numbers on the V700 actually don't look too bad considering - particularly on the higher densities.

:Niranjan.

 

[bernard_L]

Hi, Bernard....very interesting.
Just so I understand correctly - how did you calculate D from the scanner data?

In vuescan (pro version, advanced mode interface) in the Prefs tab, enable density display. Then pressing <Control> will display density values instead of RGB values at the bottom of the screen.

Also, why do you think there is such a discrepancy between preview and scan numbers - what are your resolutions for both.

For the V700, Preview: 400dpi, Scan 1600dpi. For the LS-2000, I put myself in the shoes of the naïve user: pressed "default settings" in the Input tab, then just changed from "Image" to "B/W negative" for the source.
As for the cause, I can only speculate. A proper documentation for software such as vuescan should include a signal processing diagram. Consisting of boxes (operations performed) and arrows (states of signal processing). Example of operation performed: normalize detector output to clear path (which is why the V700 film holder has a cutout at the top). Examples of states of signal: (1) black point has been set from sample of orange mask; (...) (2) processed according to scanner profile and expressed in AdobeRGB color space,(...). etc. Again, speculating, possibly Mr Hamrick does not maintain such a signal processing diagram. So, while densities should be computed as
-log10(Output/OutputClearPath),
maybe he applies the same formula to image data that is already partly processed. I have no motivation to embark into lengthy experiments to attempt to determine what is happening inside the black box just from the outside.

Preview numbers on the V700 actually don't look too bad considering - particularly on the higher densities.

One column out of 4 "does not look too bad". And maybe just changing one parameter somewhere in the interface would produce yet another column of discrepant results. Once I have calibrated my densitometer against two points on the Stouffer wedge, the agreement (my measurement versus the Stouffer data) agree within 0.02D up to 2.0D (from memory).

Then again, one can produce very nice photos, including the technical aspect, just by empirical procedures, never dealing with densities and the like.

 

[Alan Edward Klein]

Unfortunately no, IMO. Not anymore than looking at a thermometer will make the cold more bearable.

What might help you, though, is a feature of vuescan (possibly present in other software) called Multi Exposure, whereby the image is digitized twice: (a) with a moderate exposure limited by not "burning through" the transparent parts of the film, (b) second one with a longer exposure time and/or higher analog gain, to improve the signal/noise in the opaque parts; the two exposures are then combined into a single digital image with a higher dynamic range than might be achieved with either one alone. Need to store the result as 48-bit Tiff in order to reap full benefit.
View attachment 294259
Not to be confused with "Number of samples", that combines a number of identical exposures to improve signal/noise; less effective IMO.

Thanks for the info regarding densitometer.

I haven't tried multi exposures. I only use the shadow slider in Lightroom to get the most out of those areas stopping before I get noise that's noticeable.

Regarding Multi-Exposure, I have heard many claims but never saw any proof anyone supplied that differed from a single scan using a shadow slider afterward. My theory is that since a scanner is rated at a particular dMax, additional scans aren't going to produce more real data. It would be like taking two camera pictures at the same exposure settings and combining them. Will you see more in the shadows? The only way to do that is to increase exposure in a single shot. Similarly, you can't create more exposure with a scanner as it's set by its dMax which already provides the maximum light/data rating.

 

[nmp]

In vuescan (pro version, advanced mode interface) in the Prefs tab, enable density display. Then pressing <Control> will display density values instead of RGB values at the bottom of the screen.

For the V700, Preview: 400dpi, Scan 1600dpi. For the LS-2000, I put myself in the shoes of the naïve user: pressed "default settings" in the Input tab, then just changed from "Image" to "B/W negative" for the source.
As for the cause, I can only speculate. A proper documentation for software such as vuescan should include a signal processing diagram. Consisting of boxes (operations performed) and arrows (states of signal processing). Example of operation performed: normalize detector output to clear path (which is why the V700 film holder has a cutout at the top). Examples of states of signal: (1) black point has been set from sample of orange mask; (...) (2) processed according to scanner profile and expressed in AdobeRGB color space,(...). etc. Again, speculating, possibly Mr Hamrick does not maintain such a signal processing diagram. So, while densities should be computed as
-log10(Output/OutputClearPath),
maybe he applies the same formula to image data that is already partly processed. I have no motivation to embark into lengthy experiments to attempt to determine what is happening inside the black box just from the outside.

One column out of 4 "does not look too bad". And maybe just changing one parameter somewhere in the interface would produce yet another column of discrepant results. Once I have calibrated my densitometer against two points on the Stouffer wedge, the agreement (my measurement versus the Stouffer data) agree within 0.02D up to 2.0D (from memory).

Then again, one can produce very nice photos, including the technical aspect, just by empirical procedures, never dealing with densities and the like.

Thanks for the reply. I think I have the Pro version, but I can't find the Density option. Is it scanner specific - I have a different scanner, Epson Perfection 3200, which could be the reason, may be. Or may be it is the new upgrade. I have been putting off upgrading for a while. Needs more digging.

:Niranjan.

 

[nmp]

Thanks for the info regarding densitometer.

It would be like taking two camera pictures at the same exposure settings and combining them. Will you see more in the shadows?

Yes, you can. I do it all the time with 2 or more exposures.

:Niranjan.

 

[Alan Edward Klein]

Yes, you can. I do it all the time with 2 or more exposures.

:Niranjan.

Aren't you bracketing them at different exposure settings for HDR? That's different than a scanner where you can't change the exposure settings. It's fixed by its dMax designed into the machine.

 

[nmp]

In vuescan (pro version, advanced mode interface) in the Prefs tab, enable density display. Then pressing <Control> will display density values instead of RGB values at the bottom of the screen.

Found it....I wasn't looking closely enough.....

:Niranjan.

 

[bernard_L]

Edit, you found it while I was typing.

Thanks for the reply. I think I have the Pro version, but I can't find the Density option. Is it scanner specific - I have a different scanner, Epson Perfection 3200, which could be the reason, may be. Or may be it is the new upgrade. I have been putting off upgrading for a while. Needs more digging.

In my version 9.7.37 64-bit
Go to Input tab. Set Options: Professional.
Go to Prefs tab. About 60% down, tick Enable density display

then, after preview or scan, hover the pointer above the image while holding down the Control key and watch the numbers below the image frame.
This is in my version 9.7.37 64-bit (Win 10). But I've had vuescan for >15yrs, and I feel the density readout has always been here.
See also:
https://www.hamrick.com/vuescan/html/vuesc34.htm#prefsenabledensitydisplay
Anyway, from my perspective, that is useless. Unless someone can demonstrate accurate, verifiable, and consistent density readout in vuescan, following a tbd procedure.

 

[nmp]

Aren't you bracketing them at different exposure settings for HDR? That's different than a scanner where you can't change the exposure settings. It's fixed by its dMax designed into the machine.

OK. I misread your question. Never thought anyone would take two identical picture and mix them to get more DR. You can get better noise as well as mimic ND filter that way, though.

I think the point was you can increase the exposure to the max for shadows and then one for the rest. As I understand correctly, if you let it do automatically it will find an average to incorporate most of the range but if the subject has greater range then some clipping will occur. In that case, increase the exposure manually to get the shadows and then do one at lower exposure and finally combine those two in Photoshop.

:Niranjan.

Addendum:

Here is an excert from Hamrick manual:
____________________________________________________________________________________________________________________________________
Input | Multi exposure
This option provides a way to get additional detail from the darker parts of the scanned image. It is available on scanners that are able to increase the CCD exposure time.

A first pass is performed as usual with the normal RGB exposure. This will be an appropriate exposure for the image as a whole. Then a second pass is performed with a longer exposure, which can reveal additional detail in dark areas not captured in the first pass. VueScan then merges the results of the two by choosing from either the first or second exposure pass.

Professional Option: This option is displayed when the scanner can control the CCD exposure time and when scanning slide film (not Color negative, B/W negative or Microfilm).

 

[bernard_L]

Aren't you bracketing them at different exposure settings for HDR? That's different than a scanner where you can't change the exposure settings. It's fixed by its dMax designed into the machine.

Nah! Dmax is fixed (limited) by at least two things.

1. The electronic noise in the detector and analogue-digital converter; Ultimately, the limit is photon statistics.
   
2. The quantization of the signal. Assuming a very old scanner with 8-bit A/D conversion, in the darkest parts, the outcomes of the A/D converter are like 0, 1, 2; from 1 to 2 is a 100% increase: coarse! At the other end (low opacity), one can separate 254 from 255, a 0.4% increase, nice!

Note in passing that for many scanners, the spec for Dmax assumes that quantization is the only limit, when in fact the electronic noise may make the performance worse.

If the density of the dark parts is more than 2.2 (approx) this means that even if the lightest parts register as 255 (good resolution) the darkest parts will be around 1 (maybe 2, or maybe 0).
Now assume you make a second exposure with 32x more exposure time. The CCD pixels will accumulate 32x more electrons. In the light parts, they will saturate; no prblem, because we already have good date for those. The dark parts will now register around 32, and the nearby values at 31 or 33, are only a 3% difference, much better than the 100% mentioned above.
Now in the computer, compute {First exposure}+(1/32)x{Second exposure}. Do this not in 8-bit, but in 32-bit arithmetic; with a few additional tricks to discard pixels that are saturated (second exposure) or have coarse resolution (first exposure).

Got it?

 

[brbo]

My conclusion: vuescan is worthless as a densitometer. And don't get me wrong: I use vuescan for all my scans.

As I said before. Vuescan will not be your problem (if you know how to use it). The problem is your scanner's linearity. Better scanners will have proper (linear) response over a wider range.

I roughly "calibrated" Vuescan exposure on patch 2 (this is on a poor Epson 4990 flatbed):











I my book, staying within 0.1D delta is anything but useless. And that is on a low flatbed.

 

[Alan Edward Klein]

Nah! Dmax is fixed (limited) by at least two things.

1. The electronic noise in the detector and analogue-digital converter; Ultimately, the limit is photon statistics.
   
2. The quantization of the signal. Assuming a very old scanner with 8-bit A/D conversion, in the darkest parts, the outcomes of the A/D converter are like 0, 1, 2; from 1 to 2 is a 100% increase: coarse! At the other end (low opacity), one can separate 254 from 255, a 0.4% increase, nice!

Note in passing that for many scanners, the spec for Dmax assumes that quantization is the only limit, when in fact the electronic noise may make the performance worse.

If the density of the dark parts is more than 2.2 (approx) this means that even if the lightest parts register as 255 (good resolution) the darkest parts will be around 1 (maybe 2, or maybe 0).
Now assume you make a second exposure with 32x more exposure time. The CCD pixels will accumulate 32x more electrons. In the light parts, they will saturate; no prblem, because we already have good date for those. The dark parts will now register around 32, and the nearby values at 31 or 33, are only a 3% difference, much better than the 100% mentioned above.
Now in the computer, compute {First exposure}+(1/32)x{Second exposure}. Do this not in 8-bit, but in 32-bit arithmetic; with a few additional tricks to discard pixels that are saturated (second exposure) or have coarse resolution (first exposure).

Got it?

You've explained it theoretically. The proof is in the pudding. I've never seen a multi-scan that someone presented in a forum that was better than a normal single scan adjusted with the shadow slider. Or the results were so slight, that it was;t worth the bother to scan multiple times and add them together, creating other issues they have to deal with.

Also regarding slowing down the scan speed to accumulate more photons. Doesn't increasing scan time only get you more noise in much the same way we get so much noise in night time cellphone pictures? We see this problem as we raise the ISO settings on a digital camera. More noise. The scanner's sensor similarly is over-driven producing noise. One has to assume that the manufacturer designed their scanner to scan at a speed that will produce the cleanest signals just as a digital camera produces the cleanest signals at it's optimal ISO, let's say 100. Otherwise, the manufacturer would be foolish not to slow it down a little and raise the Dmax specifications to get better results and sell the machine at a higher price. So if we artificially reduce the design speed, all we're doing is adding noise. IMO.

 

[brbo]

Doesn't increasing scan time only get you more noise in much the same way we get so much noise in night time cellphone pictures? We see this problem as we raise the ISO settings on a digital camera. More noise. The scanner's sensor similarly is over-driven producing noise. One has to assume that the manufacturer designed their scanner to scan at a speed that will produce the cleanest signals just as a digital camera produces the cleanest signals at it's optimal ISO, let's say 100. Otherwise, the manufacturer would be foolish not to slow it down a little and raise the Dmax specifications to get better results and sell the machine at a higher price. So if we artificially reduce the design speed, all we're doing is adding noise. IMO.

DMax and DRange are two different things. You can increase DMax with longer exposure, but the DRange will be unaffected (longer exposure will get you higher DMax but also higher DMin (you will blow out highlights if we presume you are scanning slides)). You can only increase DRange by combining scans with two (or more) different exposures.

BTW, phones (with enough computing power) make great use of multi exposure at low light.

 

[bernard_L]

Doesn't increasing scan time only get you more noise in much the same way we get so much noise in night time cellphone pictures?

Noise in night-time digital pictures comes from increasing the ISO, either increasing the analogue gain or boosting the digital output (equivalent to sliding the right slider to the left, JUST BECAUSE you cannot increase the exposure time for fear of blurry pictures. Very long exposures (minute and more) suffer from another problem: unequal dark currents in pixels.

The scanner's sensor similarly is over-driven producing noise.

Over-driven detector results in saturation, i.e. all high values registering the same maximum possible digital output. Not "noise".

One has to assume that the manufacturer designed their scanner to scan at a speed that will produce the cleanest signals

Once the number of bits of the A/D is chosen, say 12 in a middle-spec scanner, the D range in a single frame, single exposure, cannot exceed a limit, Delta-D=3.6 for 12 bits. Any slide approaching (say Dmax 3) this, and that is not clipped in highlights, will display posterization in the dark parts. Scanning speed is another issue: intensity of light source, size of CCD pixels, etc.

IMO.

Can't argue with that

 

[Alan Edward Klein]

Thanks for both of your responses. But it's not clear to me. Will slowing scan speed add noise to the shadow areas or create other problems or not?

 

[bernard_L]

As I said before. Vuescan will not be your problem (if you know how to use it).

How do you explain that same piece of film, same scanner, vuescan gives me readings differing by almost 1.0 in D among themselves, and from a proper densitometer? Sounds to me more than a linearity issue.
Your explanations in that respect would be quite helpful.

I my book, staying within 0.1D delta is anything but useless.

Meaning a G-bar of 0.615 (0.8/1.3) might as well be 0.69 (0.9/1.3) or 0.54 (0.7/1.3) ?

 

[brbo]

How do you explain that same piece of film, same scanner, vuescan gives me readings differing by almost 1.0 in D among themselves, and from a proper densitometer? Sounds to me more than a linearity issue.
Your explanations in that respect would be quite helpful.

I can only explain it by assuming that you don't fix the exposure at appropriate levels for each individual scanner. I have no idea why you wouldn't do that. It's essential. As essential as calibrating a densitometer.

Meaning a G-bar of 0.615 (0.8/1.3) might as well be 0.69 (0.9/1.3) or 0.54 (0.7/1.3) ?

No. I went to trouble to post pictures of sampling each of 5 areas on calibration target. You'll have to excuse me for not saving hundreds of screenshots to demonstrate that readings will vary very little (probably in the range of a good densitometer) across different readings, but will only be accurate up to 0.1D (and that is only on the linear part of the scanner). You could simply plot your scanner's error and use that to get the adjusted densities which will then be much more accurate that those displayed in Vuescan. Again, this is because most scanners don't have linear response across their entire range and you can't "calibrate" Vuescan at two or more exposure points as you can a proper densitometer.

 

[brbo]

Will slowing scan speed add noise to the shadow areas

No it will not. It will DECREASE shadow noise*.

or create other problems or not?

Yes it will. It will move DMin point up. So to mitigate that problem you combine two exposures together to get the best of both exposures.

* This is all assuming positive material, of course.

 

[Alan Edward Klein]

Does anyone have a multiscan combined result with a separate normal single scan result so we can compare the two?

 

[nmp]

Does anyone have a multiscan combined result with a separate normal single scan result so we can compare the two?

How about doing it yourselves and proving either way.

 

[Alan Edward Klein]

How about doing it yourselves and proving either way.

I don't have Vuescan; only use Epsonscan on my V850. I don't want to waste time downloading a new program and learning it just to try someone's idea. If this is a viable procedure, certainly someone has some samples and has done a comparison they can show between multi and single-scan shots.

 

[nmp]

I don't have Vuescan; only use Epsonscan on my V850. I don't want to waste time downloading a new program and learning it just to try someone's idea. If this is a viable procedure, certainly someone has some samples and has done a comparison they can show between multi and single-scan shots.

It's up to you.

 

[MattKing]

Does anyone have a multiscan combined result with a separate normal single scan result so we can compare the two?

It would be unlikely that you would be able to see the difference once the results are uploaded on Photrio, or down-sized to fit on a 4K TV display.
If you made large prints, the differences are more likely to be visible.

 

[brbo]

If this is a viable procedure, certainly someone has some samples and has done a comparison they can show between multi and single-scan shots.

https://www.silverfast.com/highlights/multi-exposure/en.html

 

[Alan Edward Klein]

It would be unlikely that you would be able to see the difference once the results are uploaded on Photrio, or down-sized to fit on a 4K TV display.
If you made large prints, the differences are more likely to be visible.

They could email me the original files in full resolution. Maybe they can post in full resolution just a part of the image where the shadow areas are although I would prefer a full picture.