Estimating resolution lost during scanning ? [Archive]

Ron Marshall

20-Mar-2009, 19:42

sanking

20-Mar-2009, 20:08

That would be Nyquist theory but it does not relate to my testing with the AF 1950 target. When I use that target I get about 1800-2000 spi with the Epson 4990, when scanning at 4800 spi. However, with the EverSmart Pro I get about 3000 spi when scanning with the optical resolution at 3175 spi. When scanning with my Leafscan 45 I get about 4300 spi when scanning at 5080 spi.

As noted, all of this is based on the AF 1950 target.

Sandy King

Is my assumption correct that the theoretical resolution of a scanned image is equal to the inverse of: 1/(negative resolution) + 1/(scanner resolution)?

For example: for a drum scanner that will resolve 4000 dpi (80 lp/mm) and a negative having a resolution of 80 lp/mm, the resolution of the resulting scan would be 40 lp/mm?

bglick

20-Mar-2009, 20:34

Ron, the 1/R formula applies to the amount of resolution a camera system can record... mainly a lens and film....

A scanner is almost a camera system, but not quite. If you changed your equation to the inverse of, 1/aerial resolution of lens + 1/resolution of target, you would be more accurate....but its not that easy... The reason is, a scanner can over-sample the target, then interpolate down to the desired resolution. The problem is, every scanner has a mysterious way of producing a file....its all hidden in firmware, so we can't see its procedures. But in general, a scanners capability is tied to the MTF of the aerial resolution of the lens and the quality / count of the recording media (sensor). Just like digital cameras, not all pixels are created equal... This varies greatly between scanners...

So what it really comes down to is a scanners efficiency, i.e. how much do you have to over-sample to capture all the information on the target. The lower the rez of the target, the higher the efficiency of all scanners. However, when the target rez becomes greater (relative to the ppi of the scanner), this is where better scanners separate themselves from consumer grade scanners. The only way to determine this, is to test the scanner you are considering. So often, one price you pay for lower quality scanner is bigger files to capture the same target vs. a higher quality scanner... of course, there is other reasons as well.... speed, noise, D Range, etc...

Ron Marshall

20-Mar-2009, 20:38

That would be Nyquist theory but it does not relate to my testing with the AF 1950 target. When I use that target I get about 1800-2000 spi with the Epson 4990, when scanning at 4800 spi. However, with the EverSmart Pro I get about 3000 spi when scanning with the optical resolution at 3175 spi. When scanning with my Leafscan 45 I get about 4300 spi when scanning at 5080 spi.

As noted, all of this is based on the AF 1950 target.

Sandy King

Thanks Sandy. I'm encouraged by your scanning results; much of the film resolution is being captured (with the better scanners of course).

Is the following reasonable?:

Tests of the Mamiya 7 80mm show about 120 lp/mm on film. If a scan can capture 80% of that resolution (perhaps a bit optimistic), then that gives about 96 lp/mm; which would permit a 14x enlargement, (that could be viewed at a close distance).

bglick

20-Mar-2009, 20:47

Ron, M7 recording 120 lp/mm is a very high goal.... its feasible, but only under the BEST circumstances, shooting near wide open, great lens sample, the best resolving B&W film, AND shooting a flat target..... so be cautioned with that figure to start with.....

Now, 120 lp/mm is only 6120 ppi.... (120*2*25.4) yes, this could easily be captured by a 8000 dpi scanner such as Howtek, ICG.... as well as one or two of the very best flat beds.... I think IQ3 might be the only one, but I don't recall for sure...

Ron Marshall

20-Mar-2009, 20:57

Yes I agree, that figure would be achievable only under optimal conditions. In the tests by Chris Perez et al,

(http://www.hevanet.com/cperez/MF_testing.html)

that was their result at f5.6, and even then the edge was softer. I usually shoot around f11 with mine.

With that size of neg I would be happy with 70 lp/mm from a scan, which still permits an excellent 24x30 print.

Ron, M7 recording 120 lp/mm is a very high goal.... its feasible, but only under the BEST circumstances, shooting near wide open, great lens sample, the best resolving B&W film, AND shooting a flat target..... so be cautioned with that figure to start with.....

Now, 120 lp/mm is only 6120 ppi.... (120*2*25.4) yes, this could easily be captured by a 8000 dpi scanner such as Howtek, ICG.... as well as one or two of the very best flat beds.... I think IQ3 might be the only one, but I don't recall for sure...

sanking

20-Mar-2009, 20:59

Ron,

I have tested all of my Mamiya 7 lens and assume resolution on average of about 80 l/mm with optimum exposure, development and scanning. That allows image size of about 36X44", with at least 5 lines/mm. That is pretty close to your estimate of 14X.

Sandy King

Thanks Sandy. I'm encouraged by your scanning results; much of the film resolution is being captured (with the better scanners of course).

Is the following reasonable?:

Tests of the Mamiya 7 80mm show about 120 lp/mm on film. If a scan can capture 80% of that resolution (perhaps a bit optimistic), then that gives about 96 lp/mm; which would permit a 14x enlargement, (that could be viewed at a close distance).

Ron Marshall

20-Mar-2009, 21:09

Ron,

I have tested all of my Mamiya 7 lens and assume resolution on average of about 80 l/mm with optimum exposure, development and scanning. That allows image size of about 36X44", with at least 5 lines/mm. That is pretty close to your estimate of 14X.

BTW, I agree with blglick. 120 lines/mm is possible, but very difficult to achieve in practice. 60-80 lines/mm is much more practical.

Sandy King

If you have the 43mm, how does it compare, in terms of resolution to the 80mm? I know this is splitting hairs, since all of the lenses for the 7 are excellent.

sanking

20-Mar-2009, 21:13

Ron,

I had the 43mm but sold it to a friend, and now have the 50mm. Both the 43mm and 50mm are superb lenses and capable of over 85 lines/mm.

My 80mm lens, on the other hand, is an under achiever in that the best I can get out of it is about 70 lines/mm.

I also have the 65mm, and it is a great performer, on par with the 43 and 50.

Sandy

If you have the 43mm, how does it compare, in terms of resolution to the 80mm? I know this is splitting hairs, since all of the lenses for the 7 are excellent.

Ron Marshall

20-Mar-2009, 21:26

I've been going back and forth between the 43 and the 50 for a couple of months now, but since the weather is getting better I'll have to make up my mind.

What was the reason you opted for the 50 over the 43?

Ron,

I had the 43mm but sold it to a friend, and now have the 50mm. Both the 43mm and 50mm are superb lenses and capable of over 85 lines/mm.

My 80mm lens, on the other hand, is an under achiever in that the best I can get out of it is about 70 lines/mm.

I also have the 65mm, and it is a great performer, on par with the 43 and 50.

Sandy

bglick

20-Mar-2009, 22:48

Ron, the M7 43mm is the sharpest, distortion free SWA lens made in any format, in any era. The only similar SWA lens that comes close is the SSXL 150 on 8x10. The digitar 47mm is equally as sharp, but its not quite as wide....

80 l/mm = 40 lp/mm..... not so great.... but it could have been a typo, who knows....

The SWA 43mm is almost as sharp as the 80mm, probably 85%... I use the ND center filter with the 43mm... not mandatory, but desirable IMO.... makes ready to print images at all times.....

Ron Marshall

20-Mar-2009, 23:13

I'm sure I would be perfectly happy with either of these lenses. But some weeks I think the 43 would be a bit too wide and other times I think I would prefer that.

Is there a dedicated CF for the 43, or are you using one from a LF lens?

Ron, the M7 43mm is the sharpest, distortion free SWA lens made in any format, in any era. The only similar SWA lens that comes close is the SSXL 150 on 8x10. The digitar 47mm is equally as sharp, but its not quite as wide....

80 l/mm = 40 lp/mm..... not so great.... but it could have been a typo, who knows....

The SWA 43mm is almost as sharp as the 80mm, probably 85%... I use the ND center filter with the 43mm... not mandatory, but desirable IMO.... makes ready to print images at all times.....

bglick

20-Mar-2009, 23:16

If you go wider, you can always crop ;-)

I don't think Mamiya ever made a dedicated NDCF....I use Heliopons.... IIRC, 1 1/3 stop... pretty significant if you shoot chromes....for negs, don't bother

Ben Syverson

20-Mar-2009, 23:38

120 makes you bend over backward. Who cares if you can capture 120 lp/mm if you can't scan it?

I'd rather capture 40 lp/mm on 4x5 than 120 lp/mm on 120, because I can actually scan the 4x5.

sanking

21-Mar-2009, 07:24

I used the 43mm extensively on several long trips and eventually concluded that it was just a bit too wide for my way of seeing so I sold it and got the 50mm. It suits me better but this is personal and you might find otherwise. The 50mm lens does have one small advantage in that the accessory finder is not absolutely necessary. When you put the 50mm on the body the frame lines are for the 65mm lens, but the entire viewing area corresponds almost exactly to the field of view of the 50mm lens.

Sandy

I'm sure I would be perfectly happy with either of these lenses. But some weeks I think the 43 would be a bit too wide and other times I think I would prefer that.

Is there a dedicated CF for the 43, or are you using one from a LF lens?

Ron Marshall

21-Mar-2009, 09:11

120 makes you bend over backward. Who cares if you can capture 120 lp/mm if you can't scan it?

I'd rather capture 40 lp/mm on 4x5 than 120 lp/mm on 120, because I can actually scan the 4x5.

120 drumscans just as well as 4x5 does. I definately prefer 4x5, but the Mamiya 7 is for travel and situations where 4x5 is too cumbersome or time consuming.

sanking

21-Mar-2009, 09:38

Same here. 5X7" is my favorite format but it is not practical in many travel situations so I carry the Mamiya 7 outfit, along with a Canon 50D. I use only B&W film in the Mamiya, and try to work on a tripod whenever possible.

In my opinion you can get nearly all of the detail from a Mamiya negative with scan of about 4000 spi effective. It is possible to have more information in the negative than 4000 spi but based on comparing the scans I do at 5000 spi (about 4300 spi effective) with observation with a microscope it is pretty obvious that 4000 spi gets almost all of the detail. So for most work the Nikon LS-9000 will give results as good as the drum, at least for negatives. If you are shooting transparency film the drum will definitely give better results.

Sandy

120 drumscans just as well as 4x5 does. I definately prefer 4x5, but the Mamiya 7 is for travel and situations where 4x5 is too cumbersome or time consuming.

Don Hutton

21-Mar-2009, 10:53

I just started shooting my Mamiya 7 system again after a break - the optics are awesome. A drum scanned chrome at 4000DPI is honestly as good as I could ever imagine needing. Attached is a shot I made last weekend - the shot is cropped a bit from the original (I needed a longer lens than the 150). It's Velvia 100 shot with the M7 and the 150 lens and scanned at 4000DPI with a Howtek 4500. The crop shows a 2inch square piece from a 44x30 inch print - it's a small piece of the foreground brush. I'm 100% certain that I would not have achieved a significantly better result with 4x5.

Ivan J. Eberle

21-Mar-2009, 10:54

Ektar 100 being available soon in 120 is reason enough to reconsider MF for many.

Ron Marshall

21-Mar-2009, 12:23

sanking

21-Mar-2009, 12:27

Don,

I really like the beautiful colors in the picture also. That would be a stunning print at the size you mentioned.

Sandy

Don, that's tremendous detail from that degree of enlargement. Do you recall what f-stop you used?

Robert Fisher

21-Mar-2009, 12:34

Don & Sandy, are the Schneider 110XL & 150XL lenses in the same league as your 67 lenses in terms of resolution?

Sandy, as mentioned before, my "kill ratio" (images worthy of drum scan and hanging) with my Fuji 6x9's is light years ahead of any LF camera that I have owned.

Ron Marshall

21-Mar-2009, 12:44

Robert, no they are not, but close enough, given the larger neg size with LF, and the possibility of using larger apertures with front tilt.

http://www.hevanet.com/cperez/testing.html

http://www.hevanet.com/cperez/MF_testing.html

Don & Sandy, are the Schneider 110XL & 150XL lenses in the same league as your 67 lenses in terms of resolution?

Sandy, as mentioned before, my "kill ratio" (images worthy of drum scan and hanging) with my Fuji 6x9's is light years ahead of any LF camera that I have owned.

sanking

21-Mar-2009, 12:55

In my case I would say that a much higher percentage of my 5X7" negatives are in the keeper category than would be true of MF negatives of 6X7cm or 6X9cm. However, that makes sense since I am not going to pull the trigger on a piece of large film unless I am fairly certain the negative will make a nice print. With MF it is much easier to take risks by experimenting with unusual combinations of light, subject matter, etc.

I have never used either the Schneider 110XL and 150XL lenses. I owned a 210 SSXL for a couple of years and use it on 7X17. It was a superb lens with great coverage.

Sandy King

Don & Sandy, are the Schneider 110XL & 150XL lenses in the same league as your 67 lenses in terms of resolution?

Sandy, as mentioned before, my "kill ratio" (images worthy of drum scan and hanging) with my Fuji 6x9's is light years ahead of any LF camera that I have owned.

Don Hutton

21-Mar-2009, 13:39

Don, that's tremendous detail from that degree of enlargement. Do you recall what f-stop you used?F11 and 1/15s (on a tripod).

Don Hutton

21-Mar-2009, 13:41

Don,

I really like the beautiful colors in the picture also. That would be a stunning print at the size you mentioned.

SandyThanks Sandy

I've done a 12x18 which looks great - and may well have WCI do a big print.

Leonard Evens

22-Mar-2009, 09:37

Is my assumption correct that the theoretical resolution of a scanned image is equal to the inverse of: 1/(negative resolution) + 1/(scanner resolution)?

For example: for a drum scanner that will resolve 4000 dpi (80 lp/mm) and a negative having a resolution of 80 lp/mm, the resolution of the resulting scan would be 40 lp/mm?

That is just a rule of thumb for which there is no conceptual background. Another such rule is
1/sqrt[(1/(negative resolution)^2 + (1/scanner resolution)^2]

This gives a different answer from the previous rule For example, starting with two resolutions of 100 lp/mm, your rule would give a resolution of 50 lp/mm while the square/ square root method would give a resolution of about 71 lp/mm.

In principle, you have to multiply MTFs to get an MTF for the combination.

See Jacobson's tutorial at
photo.net/learn/optics/lensTutorial#part5
and Jeff Conrad's artilce at
www.largeformatphotography.info/articles/DoFinDepth.pdf
where there is some discussion of how to combine diffraction with defocus, a somewhat related porblem.

bglick

22-Mar-2009, 10:12

Leonard.... I would suggest 1/(1/ra + 1/rb) has been remarkably accurate vs. real world testing of film and sensors. It is also what Fuji and Leica use. I am also very aware of the other formula, which I have found to be less accurate..

More importantly...have you fully read and comprehended the Conrad link you provided? It's such an involved white paper, it would take me a week or two to try to digest his math and theories, I am curious if you worked through it yourself, knowing you are a math maven :-) If so, what is your comments?

there is a few sections in his paper that give me pause.... such as his comment about MTF only providing resolution information, but no information on contrast, which he rightfully suggest is very important in the perception of resolution. But we all know, MTF represents the % of transferred contrast from target to image plane. Makes me think he must not comprehend some key optical definitions, as this is as basic as it gets. Although he certainly does have a very thorough approach to the subject...so I am not dismissing the entire paper over this issue.

Since he does not provide any "conclusions" throughout the document... I am curious if you understood what he was representing about DOF as it relates to lens design. IIRC, he goes through two examples, Asymetrical and symetrical... Do you know his suggested conclusion on this issue? i.e. how critical is lens design as it relates to DOF?

Nathan Potter

22-Mar-2009, 12:03

This sampling business is kinda interesting and I've never thought too rigorously about it. But for either of the above relationships to hold in a reasonable fashion I assume that the scanning array, with regular spacing (say 10um) is assumed to scan over a randomly spaced film image of discrete detail of say 10um. This also assumes that there are occasional details within the image that approach the 10um resolution limit and these are recorded by one sample of the scanner being coincident with that 10um image detail. However some of the scanner samples will only be coincident with a part of one 10um detail of the film and others will miss the detail altogether. Is the above formula more simply a practical solution to the sampling problem that can be more rigorously treated by some sort of percolation theory that would need to be employed at the resolution limit? The interest here is to limit the analysis to structures with dimensions at the resolution limit only. Is there any difference in result if both scanner and film have randomly spaced 10um detail rather than scanner regular and film random - maybe not?

I'm just trying to get this phenomena down to a physical reality. In a practical vein does it really matter if the scanner samples are randomly spaced?

Nate Potter, Austin TX.

Ben Syverson

22-Mar-2009, 12:35

The math is literally academic. If you want to approach this question empirically, just shoot a USAF 1951 and compare the results through a microscope with the results from your favorite scanner. Not to get "absolute" numbers, but to see the relative difference between "perfect" and what your scanner delivers.

There are just too many variables to be able to boil the resolution loss from scanning to a general purpose equation.

bglick

22-Mar-2009, 13:44

Just to be clear, my 1/R above related to recording onto film in a camera... not sure if it was misconstrued as it relates to scanners. However, it does relate to using a DSLR to image a piece of film, whereas the film becomes the target, however, the target has here has unknown contrast, making application of 1/R much more complicated. Hence why testing is the only solution, and for once, its the simplest solution.

As stated above, the number of variables within a scanner are too great to reduce to math, unless you knew the detail of all the firmware of every scanner. Over sampling solves most everything, assuring you can gather all the detail in the film... at least for the better scanners....

Also, if you want to test a scanner for empirical data, its best to use a scanner target, vs. film shot from an existing target, which is already 2nd generation.

sanking

22-Mar-2009, 14:03

Absolutely, and this is so easy to do.

This is my answer to people who insist that there is something to gain from scanning at real high resolution, say a resolution that is way beyond what the system and film can realistically record. Just compare your scan results at different resolution in spi with observations through the microscope.

One thing of which you can be sure, the scanner can not create any more detail in the scan that is not there in the negative. It won't happen, no way, no how.

Sandy King

The math is literally academic. If you want to approach this question empirically, just shoot a USAF 1951 and compare the results through a microscope with the results from your favorite scanner. Not to get "absolute" numbers, but to see the relative difference between "perfect" and what your scanner delivers.

There are just too many variables to be able to boil the resolution loss from scanning to a general purpose equation.

Ben Syverson

22-Mar-2009, 14:06

bglick, sanking: Exactly

sanking

22-Mar-2009, 14:24

I use chrome on glass AF 1950 targets for this kind of testing. The resolution of the targets I use is 228 CPM (cycles per millimeter). CPM solves all issue of confusion of l/mm, lp/mm etc.

But for best results you need to fluid mount with flatbed scanners. And it is hard to wrap the glass around a drum for drum scanning.

Sandy King

Also, if you want to test a scanner for empirical data, its best to use a scanner target, vs. film shot from an existing target, which is already 2nd generation.

PenGun

22-Mar-2009, 14:49

I use chrome on glass AF 1950 targets for this kind of testing. The resolution of the targets I use is 228 CPM (cycles per millimeter). CPM solves all issue of confusion of l/mm, lp/mm etc.

But for best results you need to fluid mount with flatbed scanners. And it is hard to wrap the glass around a drum for drum scanning.

Sandy King

My daughter works with glass. She could wrap that target but you would need a glass drum.