Home

Printing Big: Digital SLR, Medium Format Film, Scanning Techniques Evaluated

by Shayok Mukhopadhyay

January 2006

Contents

Motivation and Scope

About a year ago, when a full-frame Digital SLR, the Canon EOS 1Ds was approaching affordability (about US$ 5000 used), I was faced with the question of whether to go fully digital or get a high-end desktop scanner, viz. the Nikon Coolscan 9000 (US$ 2000) for my medium format equipment, the Bronica SQ, soon to be discontinued but available for next-to-nothing on the used market. That full-frame DSLRs had already surpassed 35mm film was widely accepted, and claims were cropping up on the web that they'd also surpassed medium format.

With the recent launch of a much more affordable full-frame DSLR, the EOS 5D, which has specs similar to that of the 1Ds (12.7 MP vs 11MP), this question will likely trouble a whole new set of photographers, so I thought it worthwhile to post my comparison of the 1Ds with Bronica SQ scans. Further motivation for this was provided by the generous loan of a wet-mount tray for the Nikon Coolscan 9000 by Michael Grecco, which allowed me to make a fairly exhaustive comparison of scanning techniques. If nothing else, this article will hopefully act as a guide to testing methodology for others.

The article is focused on overall print quality and sharpness. It does not deal with color fidelity (as that would vary widely from film to film) and does not care to match color between digital and film exposures of the same scene. Also left out is the issue of Dynamic Range, an important issue that would require comparison with both positive and negative film. Neither of these issues is specific to big prints.


Part I. Bronica SQ (medium format transparency film) + Nikon Coolscan 9000 vs EOS 1Ds (full-frame DSLR)

Apples to Apples: Print Size

Any reasonable comparison has to settle on a target print-size. I love shooting square images with my Bronica SQ, printing them big, and framing them with a 3-inch overmat on all sides. Since 32"x40" is the largest matboard I'm man enough to handle, this restricts me to a 26"x26" print. I compared square as well as a 2:3 (35mm proportions) images from the two cameras, printed to 26" on the long side.

For a chart of equivalent focal-lengths across formats, see this.

Test Environment

Scene 1: Bronica SQ-A with 150/4 cropped to 2:3 vs EOS 1Ds with 100/2

Here is the scene we're dealing with, as captured on the 1Ds.

Each file was sized to 17.33"x26"; the Bronica file had to be cropped as well to get the 2:3 proportions. A roughly 3"x3" section of each image with no change except for some hue/saturation adjustment to give them a similar feel and resizing for convenience of presentation is shown below. Larger sections with no resizing are available here.

Bronica SQ EOS 1Ds

What are the important distinctions between the two files?

The Prints

To make the prints, unsharp masking was applied using Barry Haynes's "Sharpen Only Edges BH" Photoshop action (see Photoshop 7 Artistry); this creates a mask that selects the edges in the image and sharpens only these areas; the SQ file was sharpened to 400,1,0 (Amount, Radius, Threshold) whereas the 1Ds was sharpened to 100,1,0. Why use unsharp masking in a comparison of cameras? Because that's how the best prints are usually made.

Each file was left at its "native" resolution for this print-size, 330 DPI for the SQ scan and 156 DPI for the 1Ds; the printer driver was left to do its own up/down-sampling.

Here is a small scan of the prints, which are 3"x3" sections of a 17.33"x26" image. I doubt it's very useful for judging the quality of the prints, but anyway...



Here's a larger scan of the prints, but honestly, you need to use the source files, sharpen them to your taste, and print them yourself (or at least view them in an image editor) to get a real feel. My judgement is:

Scene 1: Bronica SQ-A with 250/5.6 full-frame vs EOS 1Ds with 100/2 cropped square

This seems like a comparison so patently tilted against the EOS 1Ds (cropping out a third of that puny sensor) that it's almost not worth doing, so I'll merely summarize the results: Needless to say that for one who shoots square images often, this is a relevant comparison.

Scene 2: Bronica SQ-A with 80/2.8 cropped to 2:3 vs EOS 1Ds with 50/1.4

Poor 1Ds performance on Traffic Signs

Here's the scene as captured on the Bronica, cropped to 2:3, with Levels applied to brighen it a bit.



Each file was sized to 17.33"x26"; the SQ file had to be cropped as well to get the 2:3 proportions. A roughly 3"x3" section of each image (circled above) with no change except for some hue/saturation adjustment to give them a similar feel and resizing of the SQ file for convenience of presentation is shown below. The actual SQ file with no resizing and minimum jpeg compression is available here.

Bronica SQ EOS 1Ds
It may or may not be obvious from the jpegs above (it is in the print), but something is not quite right with the traffic signs in the 1Ds files: in the Hamilton Ave sign, the white of the letters look "polluted" by the surrounding green, while the DO NOT ENTER sign is strangely garbled. Here are further magnified verisions of the above files to clarify the issue:
Bronica SQ EOS 1Ds

The Prints

The objective of these prints is to show that the 1Ds's poor performance on the signs is actually evident at the 17.33"x26" size; sharpening only makes the issue worse, therefore it was not applied here.

Here is a small scan of the prints, which are 3"x3" sections of the 17.33"x26" image.

The defects should be obvious here, but here's a bigger scan anyway.

Conclusion

At the 17.33"x26" print size, the EOS 1Ds files have a subjectively better feel (which could translate to better prints), but come up short against some objective "hard" measures: high-frequency lines, traffic signs, etc. I would say that such features may well be absent from a lot of landscape scenes, so if one can avoid them, this may be a non-issue. It is obvious that as the print size increases to 26"x26", the 1Ds will be outperformed by the Bronica scans. This, coupled with the fact that a pair of 1Ds cameras (one for backup) is several times more expensive than a pair of Bronicas plus a Nikon Coolscan 9000, made me go the Bronica way. The next part of the article is devoted to getting the best scans from the Bronica.


Part II. Optimal Scanning of medium format film with the Nikon Coolscan 9000

There are many different options available for scanning with the Nikon Coolscan 9000, e.g. We shall examine these one by one to arrive at an optimal (i.e. practical with respect to time and effort yet high quality) scanning technique. As before, the benchmark is 26"x26" prints from 6x6 film, so all dimensions refered to below are sections from 26"x26".

We're using the same piece of film as in the last test, always scanned at 4000 DPI with auto-exposure switched off, shown here uncropped with no Levels adjustment.

Glass Carrier

My 120 film always comes from the lab with a slight curve. I wanted to determine if this curve was big enough to affect edge-to-edge sharpness or whether the depth of field of the scanner lens was adequate to handle the curve. And if I did use a glass carrier, would I sacrifice sharpness because of the extra surfaces involved? I made 3 different scans (all with ICE off):
  1. No glass carrier, letting the scanner choose a focus point, which I assumed would be somewhere near the center of the film.
  2. No glass carrier, but forcing the scanner to focus on the bottom right corner of the film
  3. Glass carrier, again letting the scanner choose a focus point.
Here's a comparison of an 1"x0.5" area (relative to a 26"x26" print) in the middle-third of the film from scans 1 and 3. To my eye, they're equally sharp.
No Glass, Center Section, Auto Focus Glass Carrier, Center Section, Auto Focus
Does the glass scan look a little darker? More on that later.

Now let's see a 1"x1" section from the lower right edge of the same two scans:

No Glass, Bottom Right Section, Auto Focus Glass Carrier, Bottom Right Section, Auto Focus
Big difference, huh? Let's flip this around, and force the scanner to focus on the bottom-right hand section of the film without a glass carrier (scan 2 above). Much better:
No Glass, Bottom Right Section, Focus on Bottom Right
But then, we lost the center!
No Glass, Center Section, Focus on Bottom Right

Looks like the glass carrier is a good idea after all for my film if I want to have edge-to-edge sharpness. But do the glass scans look a little darker? I think so; I'd turned off auto-exposure, and I'm guessing some light gets absorbed by the glass.

Digital ICE

ICE is an automatic dust and scratch detection and removal technology that — it is claimed by the patent-holder — works without reducing the sharpness of the image. It is available on several scanners, among them the Nikon Coolscan 9000 and Minolta Dimage Multi Pro which too handles 120 film. Using a glass carrier (as I prefer to) makes one particularly vulnerable to dust. It would be extremely valuable to determine if the technology works as advertised. (Note: ICE is not available for use on traditional, i.e. non-chromogenic, B&W film.)

I chose a 3"x3" area at the top-left corner of the film and scanned it with a glass carrier with ICE switched off, set to Normal, and then Fine. The scans were downsized by 1/3 in each dimension for convenience of presentation here, but you can see the full scans (with minimum jpeg compression) here.

Glass Carrier, ICE off Glass Carrier, ICE normal Glass Carrier, ICE fine

To my eye, there's no loss of sharpness due to ICE; it seems some of the bigger pieces of dust needed ICE to be cranked up to Fine to be completely removed. This is not usually my experience; I suspect these more stubborn dust particles were acquired by the film during my wet-mounting experiments.

Frankly, ICE is better than sliced bread. Anyone who's spent hours touching up a scan, made a big, expensive print, and gone back to spend more time cleaning the scan will know what I mean. It makes scanning big film in general, and scanning with a glass carrier in particular, actually fun.

Wet Mounting

Wet Mounting keeps the film flat, just like a Glass Carrier, but without the danger of Newton's Rings that happen when the glass is not totally, uniformly flush (at a level of micrometers) against the film. I myself have ever observed only one case of Newton's Rings among all the scans I've made, but maybe I've just been lucky.
It may have something to do with the thickness of the emulsion on the film. Avoiding Newton's Rings (while ensuring edge-to-edge sharpness) is the primary reason to use wet-mounting.

How does it work?

Nikon does not sell a wet-mounting holder for the Coolscan 9000. You have to use third-party products, one of them being the Cachet Image Mechanics Fluid Mount Holder developed by Michael Grecco, who loaned me a sample to try out. It consists of a modified Nikon holder having a single glass plate; you put a few drops of Kami mounting fluid on the plate, put the film over that, then put a few drops of fluid on the film, and put a transparent acetate sheet on the film. Surface tension causes the wet surfaces to stick to each other totally flat. There will likely be bubbles in the fluid which you'll need to push out by repeatedly rubbing a soft tissue over the acetate sheet. After this, you insert the holder into the scanner and scan as usual.

How well does it work?

The quality of scans is the same as sharply focused glassless scans, which in turn are comparable to Glass Carrier scans (I say comparable because of the slight darkening effect of the glass). Of course, the wet mount scans never run the risk of Newton's Rings.
Wet Mount, ICE off Wet Mount, ICE normal

However, Wet Mounting does seem to introduce some practical issues:

Multi-Sampling

Orthogonal to the above three options discussed is one of multi-sampling, where the scanner presumably scans the film several times, and combines the values obtained to reduce random noise, while significantly increasing scanning time. All the previous tests were done with one-pass sampling.

Two scans were made, one at 1x sampling, and another at 8x sampling (there a super-fine 16x that I didnt test), and the midtones were shifted to unrealistic levels to expose noise in the shadows so that the overall scene would look like:

Here is a comparison of a 2.5"x2.5" section of a dark area of the image from the two (brightened as above) scans, resized for convenience of presentation. The actual files (with minimum jpeg compression) are here.

1x Sampling 8x Sampling

On the actual files, the noise in the 1x sampling seems a little "busier", but it's not a difference that would translate to a print, I think.

Conclusion

I've settled on the Nikon FH-869G Glass Carrier with ICE set to normal and 1x sampling as my standard scanning technique with the Nikon Coolscan 9000 for medium format film. If you're particularly susceptible to Newton's Rings with the glass carrier, wet mounting is your only alternative, but otherwise, it's not quite worth the trouble.