Maximum *useful* resolution from a Scan

You hear about dedicated film scanners vs. flatbeds vs. drum scanners, and of course the Hassy/Imacon hybrid scanner.

Considering roll film *exclusively*. Isn't there a (scanning resolution) point at which you're doing little more then polishing grain?

I saw a comparison (the url eludes me) where a 35mm frame of film was scanned in an Epson v750, A Nikon LS9000 and Drum Scanner. Believe it or not, it's actually hard to tell the difference, at least to my eyes. Can anyone shed some light on this?

(again - *practical* difference - I'm sure there's an abundance in a lab setting, I'm not a lab rat and neither are most photographers).

Really tough question. And the 'net is littered with arguments in so many directions.

All I can say is that you need to try a few things yourself and come to your own conclusions. While you do that, I think you should consider that there are (at least) three or four kinds of valuable information that you are trying to get out of a negative or positive when you scan it:

(1) high frequency info, sharp details etc.
(2) lower freq. info, which has more to do with how we perceive tonality and broad transitions
(3) actual bit depth, which we discussed before
(4) if you feel that grain is an integral part of the "feel" of your image, then you might be willing to scan well into the grain to capture that in your scan.

Note that I added (4) as a peacemaking effort

So, just to repeat what I said before in other threads, you have to decide this for yourself, but these are some of the ingredients that go into the decision.

I will leave you with one final observation. Before I did drum scans, I truly believed that

scanned file size (megabytes) = how good the scan is

Drum scanning convinced me otherwise. I realized that I could create a 200mb file from my 4990 flatbed that was no better than an 8 bit ~50mb file from a drum scanner. Surprise! So... you really have to think about what that means in terms of the issues above and the issues we discussed before such as the optical limits pertinent to each type of scanning, and how interpolation creeps in when you least expect it.

Maximum

Well, I don't own a drum scanner, but I do have the V750 and the Nikon 8000.

Without going into all the numbers and theories, believe me there is a difference you can see.

If all you are scanning are small formats below 4x5, go with a dedicated film scanner. I have the V750 for tear sheets, and all formats above 6x9.

I love what the V750 does with my 8x10 color negative film. I hate what my V750 does with any 35 mm film.

I have an LS9000 - perfectly happy with it for the most part, though scans tend to look butt ugly! It looks like I might be giving up 35mm altogether so it'll just be for MF & 4x5. I'd love an Imacon

Though mind you, I heard the 750 does one heck of a job on LF negs

Hi Guys,
If its ok, I would like to add my thoughts about this subject. There was a mention of the quality of a drum scan by Keith.
My comments will be about scanning with a drum scanner, specifically our Howtek 7500.
I scan for a living so please keep that in mind when reading my comments.

When I scan a 35mm chrome, I can get about 125-150mb. There isnt much left after 150mb.
It also depends on the film and the scanner aperture. Velvia 50 produces a much better scan than an iso 400 film.
Negatives will produce more grain than chromes.

I had a chance to compare the same drum scanned 35mm chrome to a slightly smaller scan from an Imacon and the LS9000 with the Aztek wet mount.
There was a slight difference between the 3. The drum scan was sharper and smoother than both of the others but the difference is small.
If the LS9000 or the Imacon was faster, i would probably buy one just for 35mm's.

Medium format is a different game. For the max detail and resolution you need a drum scanner. Close but not quite is the best Hassy/Imacon.
From a 6x6 chrome or neg, the usable range is 300-400mb. From Velvia 50, new Portra 160nc and Pro160s, expect around 500mb.
After that, there are diminishing returns.
For 4xS, I can get about 1-1.5gb, 8x10, 2+gb.

This is probably the site iserious mentioned. It has many different scanner comparisons. http://www.terrapinphoto.com/jmdavis/
Some scanners were missing and not all of them really fall into an apples to apples comparison but all in all, its an amazing effort.
The side by side are invaluable but you have to take it with a grain of salt.
The drum scanners mentioned on the site dont represent the quality of other, newer scanners. Like the Howtek 7500, New Aztek Premiere, New ICG's, Hell 3900....
Drum scanners like the Tango have very large minimum apertures. Even at a high resolution like 4000 dpi, the scanner can only capture what is equivalent to 2500 dpi.

IanMazursky said:

Hi Guys,
If its ok, I would like to add my thoughts about this subject. There was a mention of the quality of a drum scan by Keith.
My comments will be about scanning with a drum scanner, specifically our Howtek 7500.
I scan for a living so please keep that in mind when reading my comments.

When I scan a 35mm chrome, I can get about 125-150mb. There isnt much left after 150mb.
It also depends on the film and the scanner aperture. Velvia 50 produces a much better scan than an iso 400 film.
Negatives will produce more grain than chromes.

I had a chance to compare the same drum scanned 35mm chrome to a slightly smaller scan from an Imacon and the LS9000 with the Aztek wet mount.
There was a slight difference between the 3. The drum scan was sharper and smoother than both of the others but the difference is small.
If the LS9000 or the Imacon was faster, i would probably buy one just for 35mm's.

Medium format is a different game. For the max detail and resolution you need a drum scanner. Close but not quite is the best Hassy/Imacon.
From a 6x6 chrome or neg, the usable range is 300-400mb. From Velvia 50, new Portra 160nc and Pro160s, expect around 500mb.
After that, there are diminishing returns.
For 4xS, I can get about 1-1.5gb, 8x10, 2+gb.

This is probably the site iserious mentioned. It has many different scanner comparisons. http://www.terrapinphoto.com/jmdavis/
Some scanners were missing and not all of them really fall into an apples to apples comparison but all in all, its an amazing effort.
The side by side are invaluable but you have to take it with a grain of salt.
The drum scanners mentioned on the site dont represent the quality of other, newer scanners. Like the Howtek 7500, New Aztek Premiere, New ICG's, Hell 3900....
Drum scanners like the Tango have very large minimum apertures. Even at a high resolution like 4000 dpi, the scanner can only capture what is equivalent to 2500 dpi.

Click to expand...

Your file size for optimum quality seem very sound to me, and in general agree with the notion that has previously been expressed here to the effect that the optimum file size can be determined in most cases by the optical resolution of the camera system.

You indicate that there is very little to be gained from a 35mm scan of Velvia 50 chrome with file size beyond 150 mb. That is about 4500 spi by my calculations. If one backs that out you will find that this corresponds to resolution on film in the 90 lp/mm, which is very close to the limit of the very best combination of camera/lens/film. In other words, no matter what size print you plan to make there is no point in scanning at a resolution of more than about 5000 spi since the chances of getting more than 100 lp/mm on film (any film!!) are almost nil.

Most MF equipment can not come close to 90 lp/mm. In fact, if you do your own resolution tests you will find that even 60 lp/mm is not easy to achieve. So from a practical consideration a scanning resolution of 3500 spi is enough for most MF negatives and chromes, IMO. A 6X7 chrome scanned at 3500 spi in 16 bit RGB gives a final file size of about 435 mb, which is consistent with your notes on optimum file size. The one exception to this would be negatives and chromes from Mamiya 6 and 7 systems. These systems have demonstrated the capability of putting 90-100 lp/mm on film with several lenses. No way you can capture this in scanning unless you have a scanner capable of "effective" resolution of more than 5000 spi. Very few scanners, even drums, can do this.

BTW, my comments on scanner resolution are based on the concept of *effective resolution*, not optical. We all know that an Epson V-750 can not deliver the stated optical resolution of 6500 spi, or even close.

Sandy King

Hi Sandy,

There is one benefit to scanning at a higher resolution.
The interpolation is handled with the raw data from the scanner (high bit).
Generally better than enlarging it in photoshop from my experience drum scanning.

Another is apparent film grain. Velvia is much tighter than other films. A larger scan from Velvia would show less apparent grain therefore a much smoother image.

You are correct that most scanners have a limit to there max res. My Howtek 7500 has a max res of 5k at an aperture of 10.16 or 5.08 uM. There are 1 or 2 that i know of that have better resolution, the Aztek Premiere for one.

Best
-ian

IanMazursky said:

There is one benefit to scanning at a higher resolution. The interpolation is handled with the raw data from the scanner (high bit).

Click to expand...

But software interpolation is noiseless, no? And can be done at [almost] arbitrary bit depth. Some scanners simply insert noise when they interpolate, rather than smooth transitions between "real" data.

Anyway, this is a stray point, I wouldn't disagree that if grain plays a role in the image, then the frame should of course be scanned into the grain. From a grain-aesthetic point of view, it can be fairly argued that "clean" PS-interpolated files are too clean. Especially in the b&w case.

"I don't believe it is possible to capture film grain..."

Amen to that, Sandy.

What we call Grain - in a print - is a 2 dimensional image of the 3 dimensional 'constellations' of silver particles in the film.

It is like interpreting the stars of Orion in the night sky as a 2 dimensional picture, when it is really gazillions of miles deep.

Here is a very good summary of how film works, how scanning works, and why we get what we get:

Film Grain, Resolution and Fundamental Film Particles, by Tim Vitale

http://aic.stanford.edu/sg/emg/library/pdf/vitale/2007-04-vitale-filmgrain_resolution.pdf

...

df cardwell said:

Here is a very good summary of how film works, how scanning works, and why we get what we get:

Film Grain, Resolution and Fundamental Film Particles, by Tim Vitale

http://aic.stanford.edu/sg/emg/library/pdf/vitale/2007-04-vitale-filmgrain_resolution.pdf

...

Click to expand...

Thanks for posting the link to that article. It has some fascinating information.

Sandy

The article is very nice, thanks. It makes intuitive sense.

Like I said elsewhere, if someone is very much into the role of traditional grain in their photographs then perhaps they should simply adhere 100% to the traditional wet process. I truly don't say that to poke anyone in the eye, I am just saying, we have to be realistic about what real information a scanner is actually pulling out.. per dollar.

All tools have different signatures. The limitation of what a film scanner can SEE creates the signature for scanned film. Not a big issue. Certainly not an issue in making 8x10 palladium prints from 35mm TriX !

IanMazursky said:

Hi Sandy,

There is one benefit to scanning at a higher resolution.
The interpolation is handled with the raw data from the scanner (high bit).
Generally better than enlarging it in photoshop from my experience drum scanning.

Another is apparent film grain. Velvia is much tighter than other films. A larger scan from Velvia would show less apparent grain therefore a much smoother image.


Best
-ian

Click to expand...

I posted this earlier on the LF forum but thought it belonged here as well since the thought came from Ian.

I decided to test the concept of scanning at a higher resolution than needed via interpolation and you are absolutely right about grain. I made the tests with an EverSmart Pro that has optical resolution of 3175 ppi. I made one scan at an interpolated resolution of 8000 ppi, another at the optical resolution of 3175 ppi. In Photoshop I downsized the one made at 8000 ppi to 3175 ppi and have attached the two files. I though the test would be interesting for my scanner since the analogue to digital conversion in the scan is done in 14 bit, but the files are saved in 8 bit.

The two scans used identical scanner driver parameters, and no smoothing or sharpening was done for either scan.

I think it is pretty clear that the scan made at 800o ppi and then downsized to 3175 ppp has smoother grain, and detail also appear to be quite a bit better than the scan made at the optical resolution of 3175 ppi.

The original was a 6X7 TMY negative made with a Mamiya 7II camera and 43mm lens. Since I used the camera on a tripod, and at f/5.6, I calculate that the resolution on film is between 80 - 100 lp/mm. The area shown is .1"X .1" in size, so at current image size would correspond to a print of 135" X 165".

Sandy King

Very nice and illustrative example!

Indeed, averaging over neighboring scanned pixels (which is what you did here, in a nutshell) is a great way to boost signal over noise.

I had this discussion with a certain famous someone who was saying that 5D files are much cleaner even than 4x5 velvia. I was saying, look, if I drumscan a 4x5 slide to full res and downsample to your desired megapixel rating, my file will be completely noiseless and will utterly blow you away with detail! Of course if you don't downsample then the LF file looks noisier in a 100% pixel view... but it's a simple issue of signal to noise. Anyway my message hit some egos a little too hard and the moderator for the forum deleted it. lol

The grain is indeed less apparent, but the edges of the object are a little softer too. If I ever get a decent scanner, I'll have to try this and see what I think. Working on buying a drum scanner, but not sure I can really afford it (working on a side deal to try to find a way).

Greg_E said:

The grain is indeed less apparent, but the edges of the object are a little softer too. If I ever get a decent scanner, I'll have to try this and see what I think. Working on buying a drum scanner, but not sure I can really afford it (working on a side deal to try to find a way).

Click to expand...

Yes, I think you are right about the edges being a little softer. I had not noticed it earlier.

You are looking at the Howtek 6500, right? I was also thinking about bidding on it. Not sure yet if I will as I already have a pretty nice flatbed, but I could probably able to pull a bit more out of my Mamiya 7II negatives with a drum scan. Anyway, good luck if you decide to bid.

Sandy

Did you use bi-cubic smoother or bi-cubic sharper when you downsampled? If you did the smoother, try the sharper and see what happens, might gain the edges back without the grain.

Greg_E said:

Did you use bi-cubic smoother or bi-cubic sharper when you downsampled? If you did the smoother, try the sharper and see what happens, might gain the edges back without the grain.

Click to expand...

Greg,

I used the default Bicubic (best for smooth gradients). No doubt bicubic sharper would be better for this reduction.

Sandy

Optimal scanning resolution

To clarify this discussion about a useful resolution a bit, let's give my two cent's worth.

First of, I see many people stating scanning resolutions in terms of MB (MegaByte) size files. This is a pretty much useless way of stating scanning resolution, as the actual file size of an image is depended on many factors, like size of the original negative / diapositive (35mm, MF, LF), the bit depth captured (1, 24, 48 bit etc.) and even file format (JPEG, TIFF, PSD).

The ONLY valid way to specify scanning resolutions is in "PPI" (Pixels Per Inch), often confused or replaced with DPI (Dots Per Inch), which is actually a way to represent printer resolution, as these use ink DOTS(!) to represent images on paper, while images themselves consist of PIXELS, hence PPI!

Now for the "useful or optimal resolution". Actually, to clarify this, I am going to introduce three scanning "categories" to which a scan can be made (note I am writing "categories", not "resolutions", as the actual PPI resolutions will depend on film type). Please note I will use the term "grain", although I agree with some of the people here that "dye clumps" or some other term might be better for some type of films. However, since we all know what is generally meant with "grain", I will stick to that.

In addition, also note that these categories assume high quality scans, like from dedicated film scanners, Imacon or drum, scanners that actually have the optical or effective resolution (in PPI) to capture "grain". For example, my cheap-ass flatbed / filmscanner Canon 9950F is incapable of actually revealing grain in any but the most grain rich film types (400 ISO minimum), due to it's low effective resolution (about 1200 PPI). It just adds smeared blur at higher resolution...

So, what are these categories?:

1) - "Grainless scans"

This category of scans has been made at PPI resolutions where it is IMPOSSIBLE to distinguish "grain" (or at least virtually non-present...). Of course, these scans are therefore made at relatively low scanning resolutions (PPI). These images are virtually indistinguishable of digital captures using digital camera's. Scans in this category can generally stand a moderate amount of sharpening, without ruining the image.

2) - "Non-obtrusive grain"

This category of scans starts to reveal grain, however, the grain pattern is relatively smooth and therefore non-obtrusive. Actually, this is a VERY nice category to scan in, as it will reveal the true nature and source of the digital image (film), by showing it's grain, opposed to a full digital capture. Images in this category can only very mildly be sharpened, otherwise disaster will strike.

It is my preferred scanning category, especially for B/W film.

3) - "Obtrusive grain / scanning into the grain"

This category uses such very high PPI scanning resolutions, so as to reveal the grain pattern fully. "Grains" or dye clumps will be clearly visible. This category of scans therefore does NOT stand sharpening. Sharpening these images will just enhance the grain pattern, and in most cases destroy the image.

Personally, I do NOT like to scan in this third category, however, most people do as they feel the need to capture the highest PPI resolution possible from their images, in an attempt to obtain as much information as possible from their film. However, personally, I feel the captured "obtrusive grain" does not add to the quality of the scan, but I realize this is a subjective opinion.

So how to relate these "scan categories" to "optimal scanning resolutions"??

To do this, you must take into account the film type, and hence ISO. The more grain a film has, the lower the PPI resolutions for each category will be. Again, these are just guidelines based on my own experiences:

ISO 400 film:
- Grainless: 1400 PPI
- Non-obtrusive grain: 1800 PPI
- Obtrusive grain: >1800 PPI

ISO 200 film:
- Grainless: 1800 PPI
- Non-obtrusive grain: 2200 PPI
- Obtrusive grain: >2200 PPI

ISO 100 film:
- Grainless: 2400 PPI
- Non-obtrusive grain: 3000 PPI
- Obtrusive grain: >3000 PPI

ISO 50 film:
- Grainless: 3000 PPI
- Non-obtrusive grain: 3600 PPI
- Obtrusive grain: >3600 PPI

Again, these are just rough guidelines. For the same ISO, some films will have a more pronounced / bigger grain than others. So view this with a pinch of salt.

However, to illustrate this, I have included a number of actual scans below. All made on a high quality Imacon Flextight 949scanner. All samples are without any sharpening applied (accept maybe what the scanner driver software does internally, outside control of the operator).

So here are the samples, divided in the categories. Please note I DO not have scans for each ISO mentioned above, as I haven't made these high quality Imacon scans for each film speed, but it should give some idea of the categories. The "overview" images show the entire image, the others images are 1:1 (100%) detail images, so each captured pixel is truly visible. By the way, notice the incredible details captured in some of these scans. For example, notice the fine grating visible in the openings of the tower of the Duomo of Siena (image “grainless_provia100f_2040ppi_detail1.JPG”), and also the lighting conductor to the left of the domes top structure (image “grainless_provia100f_2040ppi_detail2.JPG”). These details are just a pixel or so wide, and still properly rendered! A major feat of the Imacon Flextight 949…:

************************************
GRAINLESS:
************************************
Film: Provia100F
Scanning resolution: 2040ppi

Overview:


Detail1:


Detail2:


************************************
NON OBTRUSIVE GRAIN - SAMPLE 1:
************************************
Film: Kodak320TXP
Scanning resolution: 1800ppi

Overview:


Detail1:


************************************
NON OBTRUSIVE GRAIN - SAMPLE 2:
************************************
Film: Kodak125PX
Scanning resolution: 3000ppi

Overview:


Detail1:


************************************
OBTRUSIVE GRAIN:
************************************
Film: Provia100F
Scanning resolution: 3200ppi

Overview:


Detail1:


Detail2:


So, actually, to my feeling the last samples of Provia100F at 3200ppi are "over-scanned". I much prefer the 2040ppi result of the Duomo.

Marco
My website

Just a few last remarks to my last post with the images to avoid confusion:

- For those who may have wondered if the "2040" ppi resolution was a typing error and should have been "2000" or "2400" ppi: no, it's not a typing error. For full frame 4x5 scanning, the highest resolution that can be set in the Imacon software IS "2040" ppi. Only when you crop or use a smaller 6x9 or 6x6 negative holder (as has been done with the 3200 ppi Provia100F scans), is it possible to set a higher resolution for 4x5 scans. For 35mm and MF, the Flextight scanners do go much higher (optical 6400 for the 848 and 8000 for the 949 at 35mm).
But actually, these software imposed limitations are not a real problem. A 2040 ppi scan of 4x5 already results in a highly detailed and huge scan of about 8000*10000 (so 80MP), which is more than enough for most practical applications.
- To be a bit more precise, all scans are from an Imacon Flextight 949, except the Kodak125PX, which were made with a Flextight 848.

And I also discovered this nice review of a Flextight 949 by Giorgio Trucco, which also shows some of the scanners internal technical assembly and also briefly discusses / explains some common scanner properties like dynamic range, noise and sharpness:
Dead Link Removed

Marco

Actually, after reading Giorgio Trucco's excellent article that I referred to in the last post, I have realized the "2040" ppi maximum scanning resolution of the Imacon Flextight 949 for 4x5 LF photo's is a hardware imposed limitation, not a software limitation. It is determined by the length of the Kodak CCD used to capture the images, which is about 8000 pixels total.

This means that irrespective of negative size, the images of the 949 can never be bigger than 8000 pixels width (length is of course undetermined, as the negative is pulled past the 8000 pixel CCD), or otherwise they are interpolated, not optical results.

Yup, that's correct it is limited to 8000 wide.

There are two Imacon Precision II and an 848 on ebay right now. It looks like at least $3000 for the Precisions and no bids on the 848 at a starting of over $8000.

You can get a damn nice drum scanner, refurbished from Aztec with software, for less than $8000 and be able to get at least 4000ppi optical up to 18x24 if you get a scanner that will take the 8 inch drums.

Also just a note... As I was reading this I noticed something that seems to go around a lot. He mentions that the dynamic range of an A to D circuit can be computed by it's resolution in bits with 8 bits being a dynamic range of 2.4. this may or may not be true of any A/D circuit. What is really more true is that a given 8 bit circuit will divide the maximum to the minimum voltage into 256 levels. If the CCD can handle the amount of light to measure a 5.8D range, than that will be divided into 256 levels for output. What I'm saying is that the number of bits really has nothing to do with the amount of dynamic range a device can measure and I really wish people would stop expressing it in those ways.

Look at it this way:

You have a 1 gallon bucket.

You can divide the contents (from full to empty) into the following amounts:

2 bits (half gallon)

4 bits (four quarts)

8 bits (8 pints)

16 bits (16 cups)

128 bits (128 ounces)

You are just slicing the measurement into smaller chunks.

In the way that this article represents the bit depth to dynamic range, you would never be able to sample a full bucket, or you would never be able to sample the empty bucket (depending on where you start counting).

And I stopped reading at the point where he went into bit depth equaling dynamic range because this always makes me angry. All it does is help the scanner makes present their (sometimes) false claims as being better than another maker. While making the granularity smaller is helpful in expressing very large ranges, it does nothing to saying what the maximum volume can be.

The only way that bit depth equates to dynamic range is if you use a fixed step size to measure from full output to low output, which is not how most A/D circuits work.

Greg: calm down and take a deep breath... ;-).

Actually, if you would have read on just a few more lines, you would have realized Giorgio Trucco just uses the bit depth story to illustrate the same thing you are pointing out: that scanner manufacturers misuse the "bit depth" of the A/D converter to come up with some completely nonsense measure of a supposedly "dynamic range" of a scanner. The used method of thinking and calculation is however, as you pointed out and Giorgio tries (but maybe not wholy succeeds), just completely crap.

Yes, you are perfectly right bit depth does not equate to dynamic range, and that any bit depth can be used to code any kind of density range (even a 1 bit A/D converter could code for a full dynamic range, if 0 was the MIN density, and 1 the MAX, of course, this would give crap results and precision, but hey, this is just a theoretical joke ;-).

But than I would be able to add some other issues as well. Who says the A/D converter needs to use a logarithmic function for converting the measured densities to bit values??? Essentially, an A/D converter could do anything it liked: use a linear, parabolic, sinusoidal or whatever... just what the circuit is programmed and build to do. Not that these would be of any practical use to encode densities in bit values... but theoretically, it could.

So, yes, you're completely right, but Giorgio did not intend to support the scanner manufacturers either...