How many ppi is enough for scanning 35mm?

Discussion in 'Digital Darkroom' started by alan_rockwood, Sep 18, 2011.

  1. I have often read comments in forums such as to the effect that scanning at numbers close 4000 ppi extracts all the image detail available in a 35mm image. I have long been skeptical of such comments. Here is a link to a scanner test that shows that scanning at higher numbers actually does extract more image detail.
    The authors of the test compared results from a high end Nikon Super Coolscan 9000 ED scanner with an even higher end Hasselblad Flextight X5 scanner. The Nikon has a nominal resolution of 4000 ppi with an actual resolution of 3900 ppi, and the Hasselblad had a nominal resolution of 8000 ppi with an actual resolution of between 6150 and 6900 ppi, depending on the film orientation. The test scans showed that the 6000+ bit scanner clearly provided more image detail than the 3900 bit scanner, and it wasn't just the ability to resolve grain, but actual image detail. It was not even a case of the two being so closely matched as to require careful comparison. The difference was clearly seen by casual observation of blowups of a small part of an architectural image.
  2. What are your requirements? For the absolute best results, you need to have your images scanned with a drum scanner. That's not an inexpensive proposition, but the results are second to none. Of course moving up to a larger film format also provides better detail, and scanning it with a drum scanner is better yet! How good do you need it to be?
  3. Alan, are you using a Nikon or Hassy scanner? If the answer is no and you have an Epson at hand, I dare say that anything above 2400dpi really makes much difference for 8x10 prints.
  4. This has nothing to do with spi. The Hasselblad is better because it has better lenses, film carriers, etc. The question you ask would be answered by comparing scans made at (say) 3000, 4000, and 6000 spi by the SAME scanner.
    Edit: From the site you linked:
    The respective maximum resolution of both scanners is only available in the 35-mm image format, and the Flextight X5 "only" achieves an effective value of about 6900 ppi in a scan resolution of 8000ppi. Thus, the difference in the effective resolution that is possible to achieve between the X5 and the X1 is of 6900ppi against 6150ppi. Both of these values are so extremely high that they exceed the resolution limit (grain boundary) of a 35-mm film.
    Thus, the difference in the resolution is rather nominal, as in case of the medium and the large formats, there is no difference in the resolution of both devices. Therewith, the resolution criteria can be omitted as a real differential factor and should not have any influence in the purchase decision. [emphasis mine]
  5. 300 pixels of scan for each inch of print should be sufficient.
  6. The correct answer, is more than most people can afford. :) Besides, it's so much easier and affordable to print negatives in a darkroom.
  7. I've scanned 35mm at 4000 PPI (Nikon Coolscan V) and 5400 PPI (Minolta Scan Elite 5400). 4000 PPI will get most everything, in most cases. But occasionally, fine detail will become indistinct at 4000. One instance I noticed was vent holes on my wife's shoes, when she is about 10' distant, shot with a 50mm lens. The 5400 DPI picks up the vent hole detail clearly. At 4000 DPI it's indistinct.
  8. Les,
    The comparison of interest to me was between 3900 ppi and 6000+ ppi, not between 6150 and 6900. Those numbers (3900 vs. 6000+) were based on shots of test targets, which means that any effects of lenses, carriers, etc. are already folded into the result. (By the way, the Nikon scanner reaches 97.5% of its theoretical maximum resolution, which shows that lenses, carriers, etc. are not significantly limiting on that instrument.)
    You are right that maximum resolution on the Hasselblad scanner is available only for 35mm. I didn't bring up tests on the other formats in order to avoid complicating the discussion. However, the results on medium format do not contradict the fact that scanning in the range close to 4000ppi is not quite enough to pull out the full resolution possible from film. In that case it was the Nikon that resolves slightly more than the Hasselblad, in accord with the difference in rating (3900 vs. 3200).
    Note added in editing. Mendel's results is consistent with the point that there is, at least in some cases, more resolution to be had from some film shots than can be captured using 4000ppi.
  9. What are you going to do with the scanned images?

    That makes more difference.
  10. I would like to be able scan without leaving any image quality on the table. If one can achieve that then the question of "what are you going to do with the image" is moot because if you have all the image quality in your scanned file you can always back off for a specific application (posting web images for example), or use all the quality if you want to make very large enlargements, but if you haven't scanned to achieve all the image quality available then you can never get it back without starting over and making a better scan, something which may not actually be possible if the available hardware doesn't support it.
    The next best thing would of course be to scan without leaving much image quality on the table. It seems that this is where we are with currently available somewhat affordable 4000ppi scanners, and given the current market conditions in photography it seems unlikely to ever get any better. In that case we must be satisfied with "very good" rather than "best" results.
  11. I would like to be able scan without leaving any image quality on the table.
    Then you have to scan with a drum scanner. An Imacon won't cut it, neither will the Nikon. The best of the drum scanners is the Aztek Premier. Find a shop with a Premier and a good experienced operator who makes an effort, and you'll pull all the image information it's possible to get from the film, short of a contact print.
    That said, just about any drum scanner with a decent operator will do noticeably better then either the Nikon or the Imacon/Hasselblad with an equally qualified operator.
    Finally, there's a lot (a huge amount) more to a scan than simple resolution.
  12. digitaldog

    digitaldog Andrew Rodney

    Beware of simplistic values like “4000ppi is enough for this or that”. I can assure you a 2000ppi scan off a good PMT drum scanner from a gel mounted original will prove a vastly superior scan to a 4000ppi off a CCD Nikon or similar scanner. Put a crap lens on a 4x5 camera, its not going to provide the same level of quality as a superb lens on a medium format system. There is far more to all this than size (or max number of scanned pixels)!
  13. Of course you can get it back. A negative doesn't self destruct on scan. Scan it again with a drum scanner if you like it
    enough to print large. Simple. There's no table.

    Simple experiment: Go through the archives of images here and pick out a scan from a $200 flatbed then pick one
    from a $10,000 Hasselblad or $4,000 Nikon scanner. If you can do that, only then is my question moot.

    Do what you like, it's your time and money.
  14. Simple experiment: Go through the archives of images here and pick out a scan from a $200 flatbed then pick one from a $10,000 Hasselblad or $4,000 Nikon scanner. If you can do that, only then is my question moot.​
    Why would I look at 60kb thumbnails to judge quality I want on print. Look at a 16x20 from a PMT and a cheap flatbed and tell me if the difference doesn't matter. Looking on screen here has nothing to do with anything.
  15. I agree with just about eveyones comments so far. Yes, you can get more detail than 4000 ppi. But, your original
    photo technique needs to have been very good to get that kind of detail. I found that a very high percentage of my
    photos were covered by 4000. And, that the quality of the Nikon 9000's lens, light source and film holders was
    sufficient. More routine photos could go to print with almost no post processing. Of course I have no illusion that the
    higher dollar systems could get more detail, more ppi, more local contrast, have better lenses and so on. But,
    unfortunately my technique, film choice, and cameras of the film era were not up to the task most of the time to
    require much more than 4000ppi. I used a minolta 5400 II and it did get more detail, but the files needed more work
    than the ones from the Nikon and the difference was kind of small in the final print.

    It is a compromise. Some are ok with Epson flat beds, some are good with the Nikons, and some can't do less than
    drum scans. It is up to your personal standards and budget.

    Me, I'm really looking at MF digital in the hopefully not too distance future. I'm personally over trying to use film, I may even send out what is left of my film library to a lo cost scan service.
  16. david_henderson


    I do exactly the opposite of what you intend. Unless I know that I'm going to print a particular image, or have it accepted by a stock agency, I'll scan it quickly and cheaply on a flatbed. That equips me to handle my requirements from at least 90% of my legacy film images. If I need to get a larger better quality scan then I'll get an Imacon scan made which costs me maybe $11 a time. So I don't need to own an expensive scanner that is going to consume a heap of time. It slows me down by about 2 weeks when the need to use or print becomes apparent, is all- and in return for that I save a heap of time behind a scanner , a lot of time on a computer since my Imacon scans come to me hand cleaned in Photoshop, and I don't need money tied up in expensive scanners.
    And indeed I'm with those who prefer scans from scanners I could never justify. My Imacon scans are 3200ppi from medium format against 4000ppi from a Nikon Coolscan but I still tend to prefer them for larger prints. As a 35mm user that compromise doesn't apply.
    Also of course we can't realistically talk about how much detail there is in a 35mm photograph in generic terms. To make it worth scanning at more than 4000 ppi has a subject variable and a photographer variable. You need the quality in the photograph and the detail in the subject to make it worthwhile.
  17. Why would I look at 60kb thumbnails to judge quality I want on print.​
    I dunno. Why would you? I wouldn't.
    Looking on screen here has nothing to do with anything.​

    For you perhaps. I would take a reasonable guess that the vast majority of scanned negatives never get printed at any size at all let alone 16x20, and at best are destined for web use. Just because you print every negative you scan at 16x20 has nothing to do with anything, when everyone else is just posting them on the web.
  18. "The difference was clearly seen by casual observation of blowups of a small part of an architectural image."
    Sounds to me like the printed equivalent of pixel peeping on-screen images at 100%. How large would the entire image be if blown up to the same level as these "small parts" of the image? And how closely would you actually view that image.
    No one I know views 40x60 prints (as an example) at arm's length. Would the difference in detail really be visible at normal viewing distances?
  19. Pixel peeping? I assume you are using that in a good way and not as an insult.
    By the way, I am a scientist by profession, and one of the things I seek in life is truth. That is a second reason why I am addressing this topic, to reverse some of the incorrect technical information that is circulating about how much scanner resolution is sufficient to capture all the detail in the best 35mm film shots, best in a technical sense, not necessarily in an aesthetic sense.
  20. Photography is an art, not a science. Don't lose sight of that fact.
  21. Yes, photography is an art. But it is also a science. Don't lose sight of that fact.
  22. Trouble is where do you draw the Alan and move up to a larger format. If 8000ppi is better than 4000ppi then 16000ppi could well have even more detail than 8000ppi. In the end though 35mm does not look that great compared to 6x7 even at an 11x14 inch print size assuming we are using similar film and decent lenses. I think I would move up to a larger film format than try to extract every last detail out of 35mm.
  23. That should read "where do you draw the line Alan".
  24. For those like Alan who are interested in the scientific aspects of photography, here is a more technical explanation of why "it never ends".
    The underlying assumption in many of these discussions is that the film image has some fixed resolution and if it is scanned with that much resolution all of the information will be captured. In fact, when multiple processes, each with limited resolution, are applied to an image (or other signal), the "blurriness" of the individual processes are added together. The fancy term is convolution. The mathematics of the convolution is not always easy to define for real situations, but a commonly used approximation is:
    b_total = b1^2 + b2^2
    where b1 and b2 are the blurriness of the individual steps, and b_total is the total effective blurriness. We can think of blurriness as the reciprocal of the resolution.
    The significance of this relationship is shown in the graph below. The three curves are drawn assuming that the effective resolution of the camera and film are 2000, 4000 and 8000 dpi. In the units that are more often used for optical resolution these correspond to 40, 80 and 160 line-pairs per mm (lppm). The horizontal axis represents the scanner resolution and the vertical axis the resolution of the final scanned image. Notice that a scan resolution of about 6000 dpi is required before the 2,000 dpi image is recorded with nearly all of its resolution, but 4000 dpi comes pretty close. For more resolution in the film image, correspondingly higher scan resolution is required.
    What is the resolution of a typical 35mm film image? This can vary greatly, of course. With a very good lens and fine-grained film, 125 lppm (6000 dpi) might be possible, but it probably doesn't happen very often. The same convolution effect applies to the lens and film resolutions, and resolution will also be reduced by focusing errors, camera shake and other factors.
    It's worth mentioning that the same thing happens when a negative is printed optically or a slide is projected. Some enlarging lenses have very high resolution, but there are plenty of other degradations to the image. I think that the common experience is that a 4000 dpi scan and inkjet printing can be just about as good as the best optical prints.
    I hope this helps,
  25. Good post David.
  26. Oops! I wrote the equation incorrectly. The correct form is:
    b_total = sqrt(b1^2 + b2^2)
  27. Yes, David's second equation is correct.
    In fact, if one takes the standard deviation as the width of point spread functions then the equation is a rigorously correct way to combine point spread functions, assuming the point spread functions are statistically independent.
  28. I should also have said that the graphs were drawn with the correct equation.
    The basic assumption is that the point spread function (the form of the blurriness) can be approximated as a Gaussian function, a "bell-shaped" curve. As Alan suggests, this will be the case if the point-spread function has an origin from random noise, but the Gaussian is also commonly used as an approximation to the point-spread function of a diffraction limited lens.
  29. Actually, David's formula is true for combining any two probability distributions, provided each of the two functions have a standard deviation. (There are some valid probability distributions that do not have a standard deviation, the Lorentzian function being the best known.)
    There is a theorem from probability theory that the standard deviations of a function that results from the convolution of two functions is equal to the square root of the sum of the squares of the standard deviations of the two combining functions. The two combining functions don't have to be Gaussian, and the two combining functions don't even need to be of the same functional form. The theorem is very powerful.
    The result does not necessarily hold for combining other measures of distribution width, such as full width at half max.
  30. I'll share some simple facts.
    1 - The Coolscan 9000 captures detail very close to its nominal resolution (4,000 dpi - full RGB) with excellent DMAX, low noise, ICE (color) and speed.
    2- High frequency and high contrast detail (e.g. tree branches, hair, small lettering, etc) on most of the films I use exceed 4,000dpi (this is measured by me not by quoting other people's work). In these cases, TMAX aproaches 6,000 dpi - Velvia 4,500 dpi - Ektar 4,200 dpi - TPan 7,000+ dpi.
    3- Most landscape pictures will show this kind of detail in some areas. In real life with my pictures.
    4- From 6x7 film + Coolscan, the resolution of the scanner becomes the limiting factor on prints larger than 30x40. I print larger than 30x40 - especially on canvas since I don't like to mirror the edges for gallery wrap.
    5- It is a silly fallacy that large fine art prints are not observed up close. Most people given the chance will dive into the detail of a large interesting print as long as the overall composition is attractive. Because of this, I evaluate pictures on both, a small 2"x2" print first and then a crop at full size before committing to the print.
    6- Scanning should always be done at the max resolution/quality possible before editing since you don't want to double up the work if you decide to print/sell. Also the film is always cleaner before putting into a sleeve.
    7- Photography has an awesome combination of art and science and cannot stand by itself with just one of the two.
  31. Those are not simple facts, they are opinions.
  32. What is not a fact Scott?
  33. I think all 7 points. They are your opinions for your work and the way you choose to do it. They are not proven substantiated generally accepted facts.
    I would agree with number 5, but never having seen an independent paper on art and viewing distances, it is not a verifiable fact, just a personal observation.
  34. I won't really argue much on these things as they are objective but I can offer you test data, films and scans personally if you are interested.
    1. In addition to the independent tests available, I have verified the Coolscan results myself. If you are interesed email me and I will share the results.
    2. Same as above and I can send you direct pictures of the film that show more resolution than the 4,000 dpi the Coolscan can capture. Or more easily, since the Coolscan tops out close to its 4,000 dpi when scanning several different films - it is clear without the necessity of a microscope that the Coolscan - not the film - is the limitation.
    3. Nature is analog and fine details like small tree branches, hair, etc also provide detail beyond 4,000 dpi.
    4. 30x40 prints from a Coolscan is about 300 dpi. You may print a scan from my website at different resolutions and observe that over 300-360 dpi, a pigment (Epson in my case) becomes the bottle neck and below 300 dpi the scan is the bottleneck.
    6. Editing twice takes additional work. Film may get additional scratches or particles from going into a sleeve.
    5 and 7. Not much to add.
  35. Ah, thats the spirit, so they are no longer categorical facts. With regards objectivity, I have argued your objectivity and results before, so please don't elaborate.
    These stated facts are your opinions. Your workflow heavily favours your techniques and results, and you have a very large pair of rose tinted glasses, often those results do not align with others opinions or results.
  36. Email me if you would like me to send you any of the film or scans for the tests.
    My results are consistent with everyone else's Coolscan results (including the OP's link ). If your results are different please post them.
    I will respond to you if you have specific questions you would like me to answer.
  37. Pixel Peeping in a good way? No, I don't think so. My point is that looking at a tiny section of an image at a magnification that no one would ever use to look at an actual photograph is an interesting exercise, but not terribly relevant. That fact that you can see additional detail at, say, a 60x enlargement doesn't mean you'll be able to see it in a 16x20 or 20x30 print.
    Science and truth are fine, but the truth isn't always relevant. A Ferrari with 600 HP will probably be faster than a similar Ferrari with "only" 500 HP. But both will exceed the performance that one can actually make use of on public roads. Similarly, if 4000 dpi contains all the detail you can see in the size of print or display you'll actually use, the extra detail captured at 6000 dpi may not be relevant "on public roads."
    MF film cameras could always out resolve 35mm film cameras, and 8x10 film out resolve MF; but for the vast majority of uses 35mm was "good enough." I suspect that for the vast majority of uses 4000 dpi is "good enough." Whether that is true or not isn't apparent based on this test.
  38. Bob, you are correct, for most applications the Coolscan 9000 is good enough. I wish it were 6,000 dpi but 4,000 dpi with very high dmax, ice and low noise for medium format for the Coolscan gives me great results at any print size.
    Would 6,000 dpi be better? Yes. Actually, in my opinion, 6,000 dpi would be just about ideal.
  39. Bob,
    So getting better quality by going to a larger format is good, but it is not good to try to get better quality by staying within a given format by going to a higher technical standard, e.g. higher resolution scans? OK, if that's the approach you want to take then fine, different strokes for different folks, but that does not invalidate those who would take a different approach.
    In any case, I started this thread to deal with a technical question, which is how many pixels are enough (in a technical sense) to capture all the detail in a 35mm scan. Some would prefer to turn the question into one of aesthetics (i.e. better resolution is not an aesthetic need), which was not the original question, or a value judgement (i.e. how good is "good enough"), which again was not the original question.
    Those are all valid issues for some people to discuss, and you are all free to discuss them if you want, but make no mistake; those side questions are off topic and probably deserve threads of their own.
  40. From a scientific point of view, best guesses by truly world recognised authorities on the subject suggest that to resolve film grain not just recognise it, thereby "leaving nothing on the table", 10,000 to 15,000 ppi is the figure needed to scan film. Anybody who wants to contradict the opinion of Bruce Fraser is either very brave and needs the background and facts to back it up, or is stupid. What that gives you is detail of the grain structure, that is needed to never leave anything behind.
    Of course currently that is an impractical figure, for so many reasons, equipment, technique, storage, file handling, costs etc etc. What it also does is ask a multitude of further questions you don't want discussed. The biggest of which for me would be, why bother? You are looking at the final few percent in an increasingly steep curve of technical difficulty and impracticality. From a technical level that is what you need to do, from almost any aesthetic, practical or useful standpoint it is not, the figure everybody else is looking for is what is needed for there use.
    So 10,000-15,000 ppi. But then I linked to a reputable source for that four days ago.
  41. Alan,
    To leave nothing on the table, its the technology that's important. PMT's are far more sensitive than CCD's. (It's Physics.) That means drum scanner. The two top scanners are the Aztek Premier and the ICG 380. The next thing you need is an experienced operator who listens to you, looks at what you are looking for and makes sure you get it. Of course you can learn to do this yourself, but it takes a little doing. To get enough from 35mm to do some real quality I'd go with the Premier any day...
    And to be totally up front, yes, I have one. I am biased, as it continues to amaze me (and my clients).
    Museum Quality Drum Scanning and Printing
  42. Alan, I posted this test to help answer your question:
  43. I've tried parallel & serially hooking up 3 azteks(ran out of money) & still can't get the clarity I need at the molecular level of the 35mm panatomic X I have from the 60's. Someone suggested contacting Stanford, but I'm afraid that with the Higgs Boson particle still unresolved it may be a futile endeavor anyway. Getting back to photography I guess for now I will stick with Minor White's idea that for technical data- the camera was faithfully used. Regards to all, Paul.
  44. Paul,
    If you will renormalize the lepton mass then it all works out fine.
  45. Alan, you are assuming that virtual particle Feynman loop effects will fix the infinity of film grain posts. From sad experience, I can predict only more divergence.
    Tom M
  46. You guys are just being silly now.
    Besides, we all know the real question is when, if ever, digital 135 will ever rival film 135, because it is quite obvious from inline posts that even 24 MP 135 format cameras don't come close to achieving the image quality I can get with a 110 film camera with a plastic lens from the early 80's.
    All in jest :)

Share This Page