Limitations of Scanner + Taking Lens

[Ken Lee]

Please correct my math. If this has been discussed elsewhere, please direct me to the discussion.

Let's say that in round numbers, an Epson flatbed can resolve around 2400 samples per inch - which is 1200 line pairs per inch. There is some dispute about this number, but for the sake of argument, let's use it.

Since there are 25.4 millimeters per inch, that scanner can resolve 1200/25 or 36 line pairs per millimeter.

Let's ignore film grain and the printer itself, just for the sake of simplicity. We'll presume perfectly grainless film, and a perfectly sharp printer.

Questions:

1) If the scanner can only resolve 36 lp/mm, then if the human eye can detect around 8-10 line pairs, does this imply that we can only enlarge a scanned image - even the scan of a perfectly sharp scanning target - by a maximum of 4x, before the image appears blurry ?

2) A very good LF lens resolves around 60 line pairs per millimeter. Longer lenses, and less impressive performers, deliver 45 line pairs per millimeter. Does this mean that the scanner whose resolution is only 36 line pairs per millimeter, can't tell the difference between a good lens and a shabby lens ? Is even the shabby lens, too good for the scanner ?

[engl]

It is not just a question of detail being resolved or not, it is also a matter of contrast. The Epson flatbeds will only resolve 36lp/mm at high contrast. The lens that resolves 60lp/mm (as say, a MTF50 measurement) is most likely going to have higher contrast at 36lp/mm than the lens that resolves 45lp/mm (MTF50).

[peter ramm]

Pretty simple, really. The system MTF is not limited by any one part of the chain. Rather, it is the product of the component MTFs. So, the degradation induced by each optic is proportional. A scanner MTF that is 0.5 of the film/lens MTF has proportionally greater effects on a 100 lp film/lens than a 25 lp film/lens.

That is why small formats with high film/lens MTFs are degraded more by low end scanners. You can always improve system performance by adding a higher resolution optic (or film) anywhere in the chain. So no, the lens is not too good for the scanner but a better scanner will create a better scan.

[Bruce Watson]

So no, the lens is not too good for the scanner but a better scanner will create a better scan.

Well said.

[Ken Lee]

A scanner MTF that is 0.5 of the film/lens MTF has proportionally greater effects on a 100 lp film/lens than a 25 lp film/lens.

I understand that effects are cumulative, like links in a chain. However...

When we test a scanner with a target, aren't there only two links in the chain ?

The first link, is a target of such high resolution as to be, for practical purposes, infinitely sharp. The second link, is the scanner.

Even with a target of infinite resolution, the scanner can only deliver 36 lp/mm. If we add a taking lens to the chain (and film grain and all the rest), things only get worse... no ? They can never get better than 36 lp/mm, right ?

[rdenney]

One thing about digital: As the spatial frequency for which MTF is being evaluated increases, there comes a point where it drops off suddenly. This boundary is smoother with film, unless the grain is a limiting factor. But even then the drop-off of MTF as spatial frequency increases has a different characteristic.

Thus, in a print size where that critical spatial frequency is big enough to see, the detail produced by the digital link in the chain will become visibly affected by that boundary.

The resolution of lenses is often reported in terms of line-pairs/millimeter, without any statement of the MTF that the lens achieves at that limit. In the tests I've seen, it appears that these values represent what I would call "minimum discernable detail", which in the little bit of testing I've done seems to correspond to a very low MTF--maybe only 10% or 20%. Thus, that 65 lpm is really a marginal performance, and on the bubble. It would not take much degradation to erase resolution at that frequency completely. And it is true that MTF is a smooth function, and therefore the system MTF at a given frequency is the product of the MTF of each of the system elements. So, let's say a lens delivers 60% MTF at 50 lpm. And let's say my Nikon scanner delivers 10% MTF at 50 lpm (above which it drops off completely), then the final scan will show 6% MTF at 50 lpm, which might be below the minimum discernable detail.

The Epson scanner seems to produce something in that range--maybe 20% based on a visual guesstimate from my Epson--at its limiting resolution before dropping off. If that limiting resolution is 36 lpm--and it's something in that range--then it would not take much effect of a poor lens to push the outcome lower than the minimum discernable detail. If you really want usable MTF at 36 lpm to support a given print size, then you need a lens that performs extremely well at that spatial frequency. A lens with a 95% MTF at 36 might only degrade the scanner's MTF from 20 to 19%. A lens with 50% MTF--still a decent performer--at that frequency might drop the final MTF to 10% at which point detail at that frequency is barely visible.

Thus, it isn't a matter of the Epson tolerating a bad lens. It's a matter of the Epson needing a very good lens at its limiting frequency, to prevent sending the marginal performance of the scanner at that frequency over the cliff. And yes, it also means that a better scanner will get more out of the lens.

In practice, most who scan with an Epson report that the see no loss of quality up to about a 4x enlargement, so your math supports that experience. I suspect most people are reporting this performance with modern lenses that will provide a pretty high MTF at 36 lpm. I'm not sure people really see as well as 8-10 lpm, so if it's really more like 5 or 6, the quality might stay quite high up to as much as 6x. I've seen 8x enlargements from Epson scans that looked very good, especially with a bit of sharpening to increase the apparent MTF at the visual limit.

I'm personally limited by my printer to 4x enlargements from the Epson, because the smallest film I scan in the Epson is 4x5. If I'm using roll film that requires a higher limiting spatial frequency, I have a better scanner.

Rick "suspecting the main advantage to a better scanner with the same limiting frequency is higher MTF at that frequency" Denney

[Lenny Eiger]

Ken, I just took a look at your site. Some very nice photos.... I must say that I didn't see any of them that required the kind of sharpness you would get from using a flash in the studio, maybe to get the condensation drops on the coke can to really pop. My guess is that the lenses you use and your film has all you need. If you stick a 10x loupe to the film and its sharp and the scan isn't then go to a better scanner, as in a drum... then you can zoom into the grain and get everything you want. Howtek's are very inexpensive these days...
I am much more interested on tonal reproduction and your images look like you spend a lot of time on that as well... When an image of mine isn't sharp, its always me... or the wind.
Cone once published a chart suggesting that his inks get more rez than you can from using color inks. (I did not test this myself so I say it with some reservation.) I might give that a try as well... More importantly, from looking at your work, I'm guessing you would really like it...

Lenny

[SURF]

2) A very good LF lens resolves around 60 line pairs per millimeter. Longer lenses, and less impressive performers, deliver 45 line pairs per millimeter. Does this mean that the scanner whose resolution is only 36 line pairs per millimeter, can't tell the difference between a good lens and a shabby lens ? Is even the shabby lens, too good for the scanner ?
Or the scanner lens is shabbier then shabby lens. In other words Epson produce very detailed shabby image with the help of always out of focus shabby own lens.

[Nathan Potter]

Coincidentally I have been looking into the performance of the V750 that I have. I mostly want to understand what this scanner is grabbing from my chromes and B&W films. I have several glass resolution masks surplused from the IC industry to play with.

I have started to scan these on the Epson using no correction and a linear curve from 0 to 256. The raw scans can then be brought up on screen in PS and greatly magnified (typically 1200%) such that the density of pixels can be measured. This technique gives a nice view of what the Epson is seeing.

First I was sort of interested in what spacial frequency is the finest that would yield an essentially clear space and a max black line without any artificial contrast enhancement in PS. This would be equivalent to a 100% MTF (well a not so rigorous interpretation I think) but maybe useful. Well scanning in RGB mode finds the cell at 21 lp/mm in one direction and only 15 lp/mm in the other. Finer spacial frequencies than this simply show reduces max. black and some density in the clear areas. This is 1067 dpi and 787 dpi respectively. Seems poor, but hey, this would be equivalent to something around 100% contrast.

As pointed out if the contrast standard is relaxed then one can claim a higher resolving power maybe up to 45 lp/mm (2300 spi) but the contrast may fall to the 10 to 20% range. I guess I'll try to measure this when I get a chance.

Also from such tests one can infer the size of a sample (pixel) as seen by the scanner lens. As I scan the various spacial frequency sets the finest that record an occasional full white 256, and black 0, with an RGB scan is 15 and 21 lp/mm cells for an assumed spot size of 24 and 32 um.

It is also useful to remember that even with a poor lens there is value in sampling accurately a blur pattern in order to replicate the characteristic of that lens.

The masks I'm using to make these measurements are chrome emulsion on glass with line widths from 1 um (500 lp/mm) on up to about 50 um. To be certain of dimensions these were checked on a Zeiss Axiotron microscope using a Leitz substage reticle. Chrome density is > log 4.0.

Also I've recalibrated the focus plane on my V750 after a year of use and found it to be 3.3 instead of 3.0 mm above the platen easily within the limit of accuracy of measurement for rigging up a wedge resolution mask.

Nate Potter, Austin TX.

[Ken Lee]

Your figure of 2300 spi is the same result that ScanDig (http://www.filmscanner.info/en/EpsonPerfectionV700Photo.html) obtained. They show the actual scan of the resolution target.

To get that 2300 spi, are you scanning at higher resolution ? wet-mounting ? In my (less rigorous) tests with an Epson 4990, I found that I once I got past 1600 spi, I got the same resolution no matter how high I set the scanner. But I just tested dry scans using a BetterScanning holder.

With the V700, I can't see any difference once I scan higher than the 2400 spi setting.

[rdenney]

... This would be equivalent to a 100% MTF (well a not so rigorous interpretation I think) but maybe useful.

If the density of your test target varies according to a sine wave, then this would indeed be 100%.

But this is really rather demanding of the system. The critical spatial frequency at 100% MTF is perhaps not that interesting a value--no lenses not even the really good ones can achieve that. I would suggest setting two values--50% MTF and 10% MTF as the critical values for "contrast" and "resolution". The spatial frequency at 50% is the smallest spatial frequency that will show good edge detail and contrast, while the other finds the minimum discernable detail.

If the 50% value aligns with about 4 or 5 line pairs/mm in a print, the print will look about as sharp as can be perceived by a viewer with an unaided eye. The scanner would need to produce 50% MTF at 16-20 lpm to support a 4x enlargement (4x5 to 16x20), or twice that to support an 8x enlargement. Your tests show that the scanner is actually much better than that, which is rather comforting.

At this moment, I'm looking at a 4x5 negative I just scanned on my V750, using Vuescan. The picture was made using FP4 processed in HC-110 dilution B. I struggled to get the scanner to see all of the density range on the negative, with the clearest parts (deepest shadows) blocking up very slightly. With a different developer, I might easily have reduced that density range to solve that problem (I can't measure density--I don't have a densitometer).

The lens I used to make the picture was an Ilex Paragon 8-1/2", and I was seeking selective focus so I used f/5.6. Not much could be expected of that old tessar design at that aperture. I'm seeing an edge spread over about 3 or 4 pixels in the focus plane (which is pretty thin indeed--I had to really hunt for it), which, at 2400 pixels/inch, would correspond to a lens performance of perhaps 12 or 15 lpm at high MTF. Probably about right for that lens used at a far larger aperture than it was designed for. It is not in any way limited by the scan, however. A 48x microscope view of the negative reveals no more detail. I have no problem with the softness of this lens in this application. It has a really old-fashioned look that complements the subject.

I can see where there was a speck of lint on the film when I made the exposure, and one assumes this is as sharp as is possible. It's edge is 2 pixels wide at very high MTF, and that corresponds about 24 lpm--probably the capability of the scanner at an MTF that makes a good appearance of contrast.

The experience and the numbers are consistent, it seems to me.

Rick "who absolutely wants the scanner to reproduce a fuzzy lens faithfully" Denney

[Nathan Potter]

Your figure of 2300 spi is the same result that photo-i (http://www.photo-i.co.uk/Reviews/interactive/Epson%20V750/page_1.htm) obtained. They show the actual scan of the resolution target.

To get that 2300 spi, are you scanning at higher resolution ? wet-mounting ? In my (less rigorous) tests with an Epson 4990, I found that I once I got past 1600 spi, I got the same resolution no matter how high I set the scanner. But I just tested dry scans using a BetterScanning holder.

With the V700, I can't see any difference once I scan higher than the 2400 spi setting.

Ken, I was scanning at 2400 spi. Chrome emulsion down and not wet scanned thus the image goes only through the Epson platen. I would assume that Epson has thought a bit about the effect of the platen and possibly done some sort of compensation to obtain the performance we see.

I think, roughly speaking, you are on track with your observation of 1600 spi although my initial observations show some marginal gains in resolution above that but one really needs to specify a contrast when speaking about this. Certainly comments about characterizing a V750 using MTF sort of data tells a much better story about performance.

Actually I was initially interested in determining the size of a capture pixel in order to understand what the machine was picking off my films. That is to say - what is the physical size of a capture pixel in RGB referenced to the film surface. It appears that 1 mil (25 um) is a reasonable number. BTW I should repeat this for using only the green channel.

Tests done some months ago using a Toppan resolution mask with scans at 1200, 2400 and 4800 spi showed no significant gain in resolution from 2400 to 4800 spi. 1200 to 2400 showed a bit but I didn't tabulate that that rigorously. In time I'll try to do that.

Nate Potter, Austin TX.

[Nathan Potter]

If the density of your test target varies according to a sine wave, then this would indeed be 100%.

But this is really rather demanding of the system. The critical spatial frequency at 100% MTF is perhaps not that interesting a value--no lenses not even the really good ones can achieve that. I would suggest setting two values--50% MTF and 10% MTF as the critical values for "contrast" and "resolution". The spatial frequency at 50% is the smallest spatial frequency that will show good edge detail and contrast, while the other finds the minimum discernable detail.

If the 50% value aligns with about 4 or 5 line pairs/mm in a print, the print will look about as sharp as can be perceived by a viewer with an unaided eye. The scanner would need to produce 50% MTF at 16-20 lpm to support a 4x enlargement (4x5 to 16x20), or twice that to support an 8x enlargement. Your tests show that the scanner is actually much better than that, which is rather comforting.

At this moment, I'm looking at a 4x5 negative I just scanned on my V750, using Vuescan. The picture was made using FP4 processed in HC-110 dilution B. I struggled to get the scanner to see all of the density range on the negative, with the clearest parts (deepest shadows) blocking up very slightly. With a different developer, I might easily have reduced that density range to solve that problem (I can't measure density--I don't have a densitometer).

The lens I used to make the picture was an Ilex Paragon 8-1/2", and I was seeking selective focus so I used f/5.6. Not much could be expected of that old tessar design at that aperture. I'm seeing an edge spread over about 3 or 4 pixels in the focus plane (which is pretty thin indeed--I had to really hunt for it), which, at 2400 pixels/inch, would correspond to a lens performance of perhaps 12 or 15 lpm at high MTF. Probably about right for that lens used at a far larger aperture than it was designed for. It is not in any way limited by the scan, however. A 48x microscope view of the negative reveals no more detail. I have no problem with the softness of this lens in this application. It has a really old-fashioned look that complements the subject.

I can see where there was a speck of lint on the film when I made the exposure, and one assumes this is as sharp as is possible. It's edge is 2 pixels wide at very high MTF, and that corresponds about 24 lpm--probably the capability of the scanner at an MTF that makes a good appearance of contrast.

The experience and the numbers are consistent, it seems to me.

Rick "who absolutely wants the scanner to reproduce a fuzzy lens faithfully" Denney


Rick, yes the better way to go about this type of characterization would be to pick a few spacial frequencies in the resolution values of interest say 100%, 75%, 50% and 10% then locate the minimum linewidth pairs which show those contrast values. What I'm not so sure about is the validity of picking the density values directly off the enlarged pixels on the computer screen. This technique is quite fundamentally different than using a scanning microdensitometer like I am familiar with. I guess I'll start thinking about this but would be very interested to know of any standards that would apply, say ASTM, even NBS, etc. that would refer to such a technique.

Lint I assume would be of some sort of hairlike origin perhaps 2 to 6 mils wide so easily imaged by a nominally 25 um spot on an Epson 750.

The masks I use are step function with line edges of TRO around or less than 100nm (0.1 um). The best of these are from Canon Lithography and are chrome on quartz. Pricey new, around $4000. Ain't surplus auctions great.

Nate Potter, Austin TX.

[rdenney]

Rick, yes the better way to go about this type of characterization would be to pick a few spacial frequencies in the resolution values of interest say 100%, 75%, 50% and 10% then locate the minimum linewidth pairs which show those contrast values. What I'm not so sure about is the validity of picking the density values directly off the enlarged pixels on the computer screen.

Anything I did using the computer screen was intended as a practical guesstimate--not intended to develop a model of performance but rather to tell me some parameters that might limit what I do. Also, it provides a reality check on the theoretical calculations.

I have done somewhat more rigorous lens testing using Norman Koren's test charts. These produce a sine wave output at a larger size, and then when photographed have a frequency high enough to be interesting. The value in his chart is that he includes a comparison strip. So, on a transparency, I can compare the blacks and whites at the test frequency with blacks and whites in a similar strip at low enough frequency to be near 100% MTF. This is imprecise but I suspect reasonably accurate for practitioners. I made the comparison using a microscope, not a computer screen. Of course, I was testing lenses not scanners.

I remember reading somewhere that the MTF 50% resolution of Velvia is 37 lpm. I have not tested it and don't know, but if it's true, that does give some context to the discussion.

Rick "wishing for time to play with this more" Denney

[Joerg Krusche]

ken,

could not find the data on photo-i which show resolution data based on test target you referred to .. on which page of the report ?

thank you !

joerg

[Ken Lee]

Sorry Joerg - You are right !

The article is on ScanDig (http://www.filmscanner.info/en/EpsonPerfectionV700Photo.html), not photo-i.

I corrected my earlier post.

[Joerg Krusche]

Ken,

thank you for the information .. very informative .. soft scan .. quite a bit of sharpening may be needed unless you like the soft image,

joerg