« Eve Arnold 19122012 | Main | Quote o' the Day »

Wednesday, 11 January 2012

Comments

Call me a contrarian, but my opinion is that DSLR slide duplication isn’t really as bad as it has been made out to be. Are there better ways of digitizing slides? No doubt, but the real question is just how much better, and an even greater question is how much better do you really need.

Just for a little background, I got into scanning 35mm slides nearly 20 years ago, my first scanner being the Nikon LS-10, and have owned many different dedicated 35mm film scanners since then, most recently a Nikon 5000. My experience is that these don’t produce scans that are nearly as good as most people think they are. The scans aren’t terrible, but they have their limitations.

I also own and use both an Imacon 343 (fake drum scanner) and Howtek D-4500 (real drum scanner), and have also used the Imacon 848 and Aztek 8000. So I’m not entirely ignorant when it comes to making and evaluating 35mm scans.

My best results have always come from the drum scanners, but there’s more to the story than that. While the drum scans are indeed “better” than DSLR dupes, they are only marginally better; it’s not a night-and-day difference.

Far more important to me is the effort required to make the scan. Even though I’m proficient at drum mounting and scanning, it still takes me at least 15 minutes to scan a single 35mm slide. I can make a DSLR dupe in 15 seconds that’s only marginally less “good” than the drum scan. Even the fastest dedicated film scanners are deathly slow compared to this.

Many specialty photographers (like me) have extensive slide collections that are entirely impossible to digitize using a drum scanner or even a film scanner like the Nikon 5000. If you’re scanning 50 slides, that’s one thing; if you’re scanning 50,000 slides, that’s quite another thing. My conclusion was that DSLR duplication was the only way I could ever digitize my collection.

I’d also like to point out that while there are many theoretical flaws to the DSLR duping technique, in practice, they are manageable and I’m not the only one who thinks so. That said, I agree that this is not a technique that’s appropriate for everyone; but to suggest that it is somehow impossible to master this technique or that this method is incapable of producing “decent” results seems overstated to me.

Interesting posts, I have learned a lot, not least my ignorance. I have been using a Nikon 60mm macro and a light tunnel from an old Noritsu printer (think minilab, it pays to have friends in the industry) backlit with a strobe. I don't feel light quality, or sharpness are problems, however I have struggled with exposure and dynamic range. HDR doesn't really work easily. I find myself making creative decisions on the mix.
However I must say even these files were better than what I could get from any flatbed scanner, or the cheap ($100) "slide digitizers" out there.
Not as good as a drum scan though....

http://www.amazon.com/Canon-65mm-1-5X-Macro-Cameras/dp/B00009XVD5

Wonder about this...as long as you have other uses for it!

Tried this with an M9 @ 160iso, a Visoflex and bellows. My El-Nikkors (50 & 80) weren't up to the task, close to 1:1, neither was any of the Elmar heads. I got best results with an old Asahi Takumar 50mm Macro @f11.

However, nothing I got approached scans from the Epson V700, @ 4800dpi, with (contrary to web advice) "medium" sharpening applied at capture.

I don’t disagree with your position on the impracticality of this idea in general, but I do disagree a little with your comments on resolution. Back in the days when people argued about film vs digital, there were lots of discussions about interpolated pixels (DSLRs) vs actual pixels (film scanners), stating that a film scanner is actually capable of higher resolution at the same pixel count. That may be technically true, but it doesn’t reflect reality as I've seen it. I own a Nikon Coolscan V (4000 dpi) scanner and I’ve made careful comparisons of scans of 35mm Fuji Velvia and Acros, shot on Zeiss prime lenses (at optimal settings, yada, yada), vs a 12 MP APS-C digital camera with a Tamron zoom. It was damn near a tie between the two formats, with regard to detail rendering. No matter how I compared the files (up-res one, down-res the other, etc.) I was never able to pull more detail from the 20.5 MP film scanner file than the 12MP DSLR file. Granted, I’m sure there are a few films, like Tmax100, that are capable of slightly more resolution than my examples. But for most purposes, I would think a 12 + MP DSLR should be capable of taking a nice macro shot of a 35mm film frame, assuming the other variables are under control.

This whole discussion makes me wonder why some company like Sigma or Tamron hasn't made a film scanning lens. Well, I guess I know the reason: it wouldn't be profitable. But, they could easily make a dedicated macro lens with a built in film holder and backlight…or even just a film holder/backlight attachment for existing macro lenses. I'd consider the "scanner lens" if it were available, because it would be a lot faster (and less noisey) than my Coolscan. I should contact the patent office. Nobody steal that idea, ok. :)

"'Anything we make that'll focus really close.' Calling a lens a macro doesn't tell you any more about its quality than labeling something an "enlarging lens.""Unfortunately, while I have tested almost every credible enlarging lens ever made, I have not done the same for macros."

It's impossible to generalize safely about lenses, especially in such specialized use.

I've done a little testing. Of the handful of true macro lenses I have, the OM Zuiko 50/3.5 and Tamron AF 90/2.8 are the best at 1:2, in resolution, flatness of field, geometric linearity and evenness of illumination. This is about the right repro ratio for camera "scans" on cropped sensor cameras.

However, the Tamron is sharpest wide open or maybe at f4, while the Zuiko is sharpest at f8. In the application at hand, where DOF is important to deal with imperfect film flatness, parallelism, etc., the old, MF Zuiko is superior.

At 1:1, for FF sensor, the Tamron is better, but the higher magnification makes DOF even more important. The Zuiko 80/4 auto bellows lens at f8-11 is my axe of choice. I don't have the range of other lenses others may have for comparison, but this one is excellent compared to anything else I've seen. No surprise, as it is optimized specifically for 1:1 reproduction by a maker with extensive experience and sterling reputation for micro/macro lenses.

In theory, at least, enlarging lenses optimized for 10:1 or so (11x14 from 35 mm) should not be at their best at 1:1.

Moose

"If you're working copying sheet film with a less-than-full-frame digital camera, the best solution is any top of the line enlarging lens, reverse mounted. "

Wouldn't reverse mounting an enlarging lens be exactly the wrong thing to do?
If you had an enlarging lens optimized for 13x magnification and were using it un-reversed on an aps c format NEX camera to photograph an 8x10 negative the lens would be in it's optimum range.

Reversed it would be optimized for photographing a 1.4mm by 1.9 mm section of the negative. Even if you were mosaicing the files that would be more than 57000 exposures to make.

I have Schneider and Rodenstock Apo enlarging lenses, and agree that they are excellent, but for high magnification--1:1 and higher, I like micro lenses. These are designed for the task, and usually come in RMS mount, requiring an adapter and bellows or extension tubes and a macro rail that moves the camera to focus. After trying a few of these, the one I've settled on is the 35mm f:/2.8 Canon FD Macrophoto, which is an RMS mount lens that comes with a Canon FD adapter, and you can easily swap it out for another adapter, like EOS or Nikon F mount. I've compared it to a couple of older Zeiss Luminars and was much surprised to find the Canon to be sharper, though newer Luminars may be better. I haven't had a chance to compare it to a Leitz Photar or the comparable Olympus micro lenses, which have excellent reputations.

The 35/2.8 in an RMS-M42 adapter and an M42-EOS adapter gives almost exactly 1:1 magnification on my Canon 5DmkII. I focus by moving the camera with a Linhof macro rail.

Reverse-mounted cine lenses for 8mm and 16mm can also be surprisingly good for this kind of work. I have some, but always come back to the Canon FD 35/2.8 Macrophoto.

Wouldn't a classical macro lens+bellows setup serve the same purpose without having to go for a reversed enlarger lens?
FWIW, the 55/3.5 micro-nikkor would be my pick for any such work: it is extremely sharp and has as flat a field as an enlarger's. With a slide viewing table as the light source, I've been able to get acceptable results with slide film - E6 process. I'm far from happy with the results for colour negative: too hard to correct the colour cast. I reckon I'll definitely need an enlarger's colour head for that...

Dear G,

"...to suggest that it is somehow impossible to master this technique or that this method is incapable of producing “decent” results seems overstated to me...."

Since I did not suggest either of those things -- if I believed them to be true I simply would not have written the columns -- I don't think it is *I* who is guilty of overstatement.

~~~~~~~

Dear Hugh,

Ah, you're entirely right, from the user point of view. Compared to its "normal" use, the enlarger lens is reversed, because you're using it as a reducing rather than a magnifying optic. But from the viewpoint of how it is mounted on the camera, it's in the normal orientation, NOT reversed-- front elements away from camera body, thread mount towards camera body.

Apologies for that confusion, my bad.

~~~~~~~~

Dear Paul,

Interesting... you were using a glass carrier with the Coolscan?

If not, who knows what resolution it was actually delivering, but I can promise it was way below 4000 ppi.

Actually, no scanning system delivers 100% of the theoretical resolution. My Minolta delivers a solid 4000 ppi from a 4800 ppi scan, and I consider myself fortunate. 85% of theoretical is damned good.

pax / Ctein

I take the point that this is a bad idea. If you're in the land of "mounting an enlarging lens backwards" you might as well just buy a good flatbed scanner. For me, the point of copying film with a digital camera is to use what you have to get what you want -- NOT to invent a new expensive and complex solution to get what you want, or something less than what you want. I have an iMac (light source), a cheap DSLR, and a macro lens. It is what it is, and it does a thing that I find non-useless.

On a completely separate note: Ctein asserts that actual resolution of a digital sensor is around 1/2 of the underlying pixel count, which I assume is due to demosaicing the Bayer array? Am I missing something, or should the resolution be closer, therefore, to the actual sensor pixel count when the view is monochromatic (e.g. b&w negatives)?

If so this suggests that you COULD build a very high quality rig based on current sensor technology which uses a bunch of tricks:

Multiple exposures across space (stitch them back together), Multiple exposures across time (super-sampling), and colored light sources to improve single-image resolution.

I'm not sure that there IS a practical limit here, and it's definitely buildable. It's just not worth building unless you're willing to build it to beat a proper scanner by quite a bit.

Caveats: I don't fully understand supersampling, and how it interacts with anti-aliasing filters (and the inherent low-pass nature of a lens with aperture, which is basically an anti-aliasing filter too), and I am unsure about my notion that a monochrome scene improves resolution on a standard Bayer array. I *think* this stuff would work, though.

Dear Andrew,

Interesting idea. But it would have to come in with a custom RAW converter. You might talk to one of the small RAW converter developers and see what they think about adding a "monochrome" switch to their software.

pax / Ctein

I am sorry Ctein, but your two articles now have convinced me that I must try the Camera method over looking at expensive scanners for my 120 roll Negatives. The set up I will now test is using an old laptop as the light source. I have Powerpoint on that laptop, so I can modify the lighting color, and I can tape or mount the negative to the monitor in center (which has even lighting). I may need to play with separation distance from monitor to avoid photographing pixels, but may not be an issue. Then for lenses I think the excellent Tamron 90mm F2.8 adaptall I have or the Olympus 35mm F3.5 Macro should be good. With all adjustments, tripod usage, filling the frame and shooting 10 images should be enough. Then load the images in to PhotoAcute, apply their super-resolution algorithm and the out comes 40Mpixel equivalent "camera scans" of my 120 roll film.

The time will not take too long once all set up. And the New Photoacute v3 has batch processing, so I could process overnight for all my rolls once shot. I think you have convinced me to at least test it out with a couple of rolls. Much much cheaper than buying even a flat bed scanner. And once all the measurements (camera distance, aperture, film location, power point slide color are worked out the first time through) following shots will be super quick.

This all started because my wife wanted to see some photos we had taken the last time we were in Rome, which was the year of the Falklands war! - I eventually discovered that I had taken slides and they were in a carousel in a desk along with my long forgotten projector. I hooked it up and after a while the fan started to get up to speed... we spent an evening going through several hundred old slides which brought back many memories.

I do have a film/slide scanner, but it is frankly very poor especially on dense slides, and only scans at 1200dpi. It is also SCSI so means I have to use a SCSI card in an old PC to even use it at all.

A few years ago I attempted to make a slide copier with my A700 plus copy stand & 100mm macro lens and a ring flash for backlighting illumination, but it was never satisfactory as focus was very critical and eventually the flash died and that was the end of that experiment.

I toyed with getting the slides copied professionally, but shuddered at a) the cost, and b) the thought of the slides going missing or being damaged.

I then realised that now that I have a full frame DSLR, the old slide duplicator attachments that we used to use on film SLR cameras might work. I had dismissed this idea before because of the 1.5 crop factor with the A700.

I managed to get a Makinon Zoom Slide Duplicator on eBay for £25. This has a T2 mount for which I already had a Minolta T ring adaptor (for my telescope).

Illumination for the duplicator was a rechargeable LED light flat panel video light. I found Electronic Flash white balance worked pretty well, though I shot in cRAW anyway. I used 3 shot bracketed at 2 stops, just in case I wanted to do any merging in PhotoMatix or Enfuse. In reality I could have got away with a single exposure, as RAW gives quite a bit of latitude.

Wish the A900 had Liveview - It would have made getting initial focus set so much easier!

How did it work out? - Very well indeed, a couple of magnitudes better than my Epson filmscan 200.

Andrew & Ctein

"You might talk to one of the small RAW converter developers and see what they think about adding a "monochrome" switch to their software.'

Consider it done, and it's free.

dcraw already has exactly that

http://www.cybercom.net/~dcoffin/dcraw/

-d Document Mode (no color, no interpolation)

better documentation can be found here
http://www.camerahacker.com/Digital/dcraw_by_example.shtml

"The -d option tells dcraw to output the image from raw data without interpolating color. The output is a gray scale image. This form of processing is faster because the process bypasses the bilinear interpolation and the VNG interpolation that are done in a normal color process. dcraw also produces a file with the ".pgm" extension, rather than the normal ".ppm" extension."


ufraw seems to be a popular front end for dcraw
http://ufraw.sourceforge.net/

and digikam which incorporates dcraw is a full blown digital photo management application like based on dcraw just like lightroom , except that it exposes some more of the guts of dcraw
http://www.digikam.org/

The only problem is that you will either have to do some work to color balance your light source and film base , or write some code to normalize the 3 color channels


Silly me,
"The only problem is that you will either have to do some work to color balance your light source and film base , or write some code to normalize the 3 color channels"

that's done too

-r Set red multiplier (default = 1.0)
-l Set blue multiplier (default = 1.0)
-b Set brightness (default = 1.0)

For digitizing 35mm slides, film flatness in a carboard mount is imperfect.
We should somehow be able to pair this with curvature of field of an imperfect 'macro' lens.
How do we do this match?
Are spec's of imperfect field flatness available? Where?
I've never researched it, but I speculate that film curvatures are somewhat consistent, or at least less than deviation from flat.
Doesn't it make more sense to complement each 'problem' than fight them individually?

Dear Hugh,

Haw! Wonderful!

Someone try this out and tell us how well it works and how much more resolution it squeezes out.

~~~~~~~~

Dear Wayne,

Kodak used to make a special lens for the Carousel projectors that had a curved "plane" of focus, for just the reason you mentioned.

In theory this is an entertaining idea, but how you'd find a lens matching the average slide curvature, I dunno.

pax / Ctein

Wayne

In the 70s Kodak researched that problem and came up with the Kodak Projection Ektanar C (curved field) Lens, supposedly after measuring the curve of a bunch of slides.

They are pretty easy to find, just about every 1970s and later slide projector came with one. I found mine sitting on the sidewalk in the trash.

I just pulled one out of my big drawer of lenses and found that using it as a relay lens* in front of my 105mm Nikkor focused at infinity it covers pretty much exactly the frame.

I think that the field curvature would work out, but I'm not inclined to put any effort into it at the moment.

I have been thinking of rigging up a carousel projector into a slide film chain as part of a project to dupe thousands of family slides using just such a setup. TV stations used to use a setup like that in the days before digital frame stores


*Either I'm focusing on the aerial image produced by the projector lens, or I'm using it as a 9.8 diopter lens depending on how you want to think about it.

Although I am OK with the results from my Coolscan IV, resolution wise, the thin depth of field gives me fits with transparencies, even remounting the film in tensioning mounts. So I have sprung for a Olympus bellows, slide copy unit and 80mm macro. Not sure if this will solve the DOF issue - at the working distance for duping, it looks like the DOF is around .02" at f8, but we'll see.

Good article, this one is better than the first one in the series. Even though I've done multiple experiments on this, it's very interesting to get ideas around the topic.

Personally I feel that alignment and flatness are the trickiest problems. I have seen focus stacking suggested, but I haven't tried it on film myself, must try to do that. It's a bit cheating, but might be the most practical solution in the end.

What comes to lenses, at 1:1 the Apo Rodagon D 75/4 is inexpensive considering the quality and a solid choice. At other magnifications good enlarging lenses or macro lenses work. The 55 mm micro Nikkors are in my experience good and of newer lenses the Zeiss macros should be worth a look. I would generally prefer a short focal length.

One more thing: the required resolution really depends on what one is scanning. For example, I tried this technique on some 100 ISO/ASA slide film from the mid 70's and the grain and detail was such that I'm confident that fully using 12 mpix is enough for that. For better resolving films (ISO 25 BW, modern color) higher pixel counts do make sense. But a friend of mine uses a Coolscan 9000 with glass holders and rarely uses the maximum resolution, since he feels that it really doesn't add anything of significance and increases the processing time too much compared to a slightly lower resolution.

For those reasons, I actually value good contrast, dynamics and low noise more than a very high resolution when scanning (I'm assuming a decent resolution of course).

The comments to this entry are closed.