Resolution Comparison of
Film vs. 6 Megapixel Digital

Version 1.1, © 2004 by Dale Cotton, all rights reserved

The Test Described

While I work with 35mm film, I occassionally re-evaluate digital capture to see if it has become a viable alternative for my own needs. My needs, as a woodland photographer, are far more demanding of fine detail than is the norm. The following evaluation is one I made recently to answer my own concerns. I present it here, not as yet another attempt at a definitive answer to the question of whether digital or film is capable of more resolution, but as a reference point for those who do similar work.

The Nikon D70 is widely renowned as having excellent resolution for a 6 megapixel dSLR. Tests show it to be very comparable to the excellent Canon 10D (and Digital Rebel) in resolution. Historically, 6 mps was considered the holy grail of digital capture. Somewhere someone popularized the idea that 6 mp, 2000 x 3000 pixels, contained the same amount of information as a 35mm film frame, minus the grain. Which type of film and which ISO film apparently was an irrelevant consideration. ;)

One thing that prompted this test was a suspicion I was nursing that the test shots I had seen from dSLR photographers were not doing their cameras justice in fine detail handling. I was seeing significant halo blurring that suggested that over-simplistic in-camera sharpening was being employed. The other thing that prompted the test was the availability of my friend's, Cristian Calineac's, Nikon D70.

My methodology was this: I positioned a tripod in front of a convenient scene with fine detail subject matter. On the tripod I first placed my usual Nikon F80 film camera with my usual 35-70/f2.8 zoom plus Reala 100 colour neg. film, and photographed the scene at 70 mm and various apertures. I then replaced the F80 with a friend's D70 using exactly the same lens but changing the focal length to 46 mm to compensate for the difference in frame size between the two cameras. An alternate approach would have been to move the tripod proportionately closer to the subject scene to retain the same field of view, instead of changing the focal length. Not only did I lack the time to do this, going closer would have changed the focal point, which I did not want to do. My approach would seem to favour the D70, since 50 mm is the sweet spot on most lenses, however my own testing shows my particular 35-70 has very equal resolution all over its range.

There was plenty of light and absolutely no wind, so conditions were good. I focused on the largest tree, about 150 feet from the camera, and the background brick building is about 300 feet from the camera. Because these distances require the so-called infinity setting on the lens, focus is eliminated as a variable. The f/8 pair of shots were used for the following comparison.

The D70 was set with in-camera sharpening turned completely off. The resulting raw .NEF files were converted to TIFF in NikonView. The Reala frames were scanned to TIFF at 4000 ppi using my Canon FS4000 film scanner. All crops below were then saved as JPEGs with very little compression (Photoshop level 10), to prevent compression artifacts interfering with your viewing pleasure.

Here is the test scene, as recorded by the D70:

Fig 1. D70 test, full frame (2000 x 3000 pixel frame)

Here is an unreduced crop from the above frame. I've applied what I consider to be an appropriate amount of sharpening. Click on this and the next figure to access larger, unsharpened crops:

Fig 2. D70 crop

As you can see, there is no sign of the grotesque blurry haloing, one sees when USM on top of in-camera sharpening is employed.

Here is the same portion of the scene as recorded on Reala; again sharpened to taste:

Fig 3. Reala crop (unretouched crop from a 3778 x 5776 pixel frame)

Note: the "unsharpened" Reala crop is actually slightly sharpened, because there is no option to turn minimal sharpening off in my scanning software. It therefore takes proportionately more USM to restore crispness to the D70 frame than the Reala frame, and we should not hold that against the D70. ;)

The Results Described

I have to pat myself on the back for choosing such a demanding scene. ;) Neither format was really up to recording the detail of the branches and twigs of these distant, leafless trees. However, the Reala scan - as unlovely as it may look on-screen - does get enough detail that acceptable prints up to at least 14x21 inches could be made, and any graininess you may see on-screen at 100 ppi does not show in a print at 300 ppi.

I tried many combinations of upsampling and sharpening, but printing the full D70 at 11x16 was the largest I could go while retaining acceptable fine detail. By 12x18, not only do the twigs and branches begin to blur, but much other detail, including faces and vehicles, which contain smooth curves, are noticeably pixelated. (One print I made was done by upsampling the D70 frame by 190%, which gives it the same dimensions as the Reala scan, then carefully sharpening that, then printing at 12x18. No joy. The twigs were unacceptably soft and the pixellation just mentioned was obvious.)

But of course, one person's tolerable is another person's wonderful. In particular, many photographers accept something called a "normal" or "typical" viewing distance, which is roughly as far from the print as its diagonal. They feel that so long as the print looks acceptable from that distance, then it is an acceptable print. I personally cannot accept a print that doesn't stand up to viewing as close as I can focus, which is about 8 inches.

For these reasons I urge you to edit and print the large raw crops for yourself. Even better: repeat the entire test yourself.

Purely in terms of resolution, what I am seeing is that the 6 mp frame is very equivalent to what I get if I scan Reala at 2000 ppi instead of 4000, which also gives me a roughly 2000 x 3000 pixel frame. (You can see this for yourself by downsampling the raw Reala crop by 50%.) The D70 frame has a wonderful creamy smoothness due to its lack of grain and digital noise. When printed at 270 ppi or higher, for print sizes of 7.4 inches or smaller, it will produce uncompromising, contact-print-like results. The less critical fine detail is to the image (or the less critical the viewer), the larger it can be enlarged: so for images such as head shots prints even up to 16x24 inches are quite practical, and most content can be printed up to the 12x18 max size of 13 inch printers with very good results.

To summarize, my own call on the D70 with pro glass is the following:

Subject matter	Close viewing distance	"Normal" viewing distance
Far-ground subject matter: distant vegetation, group portraits	Up to 11x16 (280x400)	Up to 12x18 (300x450)
General mid-ground subject matter	Up to 12x18 (300x450)	Up to 14x22 (350x550)
Near-ground subject matter, portraits, figure studies, animal close-ups	Up to 13x19 (330x480)	Up to 16x24 (400x600)

My Deductions

Based on the above analysis and, again, based purely on my own needs and observations, I see 6 mp dSLR as giving me roughly six/tenths the amount of usable detail as 100 ISO 35mm colour neg film scanned at 4000 ppi. This means to me that I could replace my F80 with a Canon 1Ds (if I could afford to); and that for me, at least, a more correct mantra would be that 10 mp of digital capture = 35mm film. However, this simplistic equation ignores the wonderful grainlessness of high-end digital capture. Put another way:

digital trades off a modest amount of sheer resolution in order to add a significant amount of image purity.

Of course, there are other critical aspects of image quality than resolution and grain. Acutance and dynamic range plus latitude are two others. I rate digital as having acutance as good as neg film and better than most slide film. And we know from other sources that the useable dynamic range of a good 12-bit dSLR RAW image file is at least 7 stops, depending on your tolerance for shadow noise. Although these 7 stops are tight on the highlight end, that's still plenty of DR for even my demands.

Finally, I just know that somewhere down the line someone is going to read this a bit too quickly then write me to ask why my conclusions differ from Michael Reichmann's and Norman Koren's. There are several factors here that come immediately to mind. Both Michael and Norman compare digital to Provia 100 slide film, which has similar resolution but dramatically poorer acutance than Reala 100 neg film. Michael scanned at 3000 ppi, where Norman and I scanned at 4000 ppi. Michael tested using an architectural subject with very few irregular or curved lines. Norman works in part at least from simulations of MTF charts. Etc. What Michael and I have in common is that both of us limit our claims to the specific tools we employ and warn that everyone needs to make his or her own tests.

(A bit of trivia: Michael's 1Ds test shot shows the Toronto Metro Hall in the background (light grey + light green with pointed roof). My test shot was done in the Metro Hall park.)

Another and unrelated fallout from the test is that I was working with the 2/3 size finder image of the D70 compared to the F80. Because I wear -3.5 dioptre eyeglasses, and in conjunction with the smaller size of the D70's finder, I found composing significantly more difficult.