A Comparison of CCD and Drum Scanners for Novices
Photography has been a hobby of mine for over two decades. Along with a professional involvement with computers, it was only natural that this hardware freak would eventually buy a scanner, the first of which was a Nikon Coolscan I. In the days before the Web and devices that could competently display the Coolscan's output, it was little used other than for recording my photography and for personal enjoyment. The Web and the Epson Stylus printer changed all that. Scanners, PCs, digital cameras, and snapshot capable color printers are now offered as mass-market items. A visit to the imaging department of a magazine publisher in New York reflects how radical changes in the mass-market can eventually drive technological developments in the high-end: off-the-shelf Macs and software have replaced specialized graphics workstations; Epson printers are being evaluated as replacements for expensive dye-sublimation printers; and networks enable rapid and widespread access of editorial material. When it comes to scanners for publishing, however, drum scanners are still the technology of choice.
If, like me, your involvement with digital scanning is as an avocation rather than as a vocation, you may have heard or read that drum scanners provide the ultimate in scanner performance. The most common technology employed in consumer scanners is the charge-coupled device (CCD). Few people, however, have had the opportunity to compare directly CCD and drum scanner output side-by-side. After a couple of friends asked me about the difference in results between the two technologies, I decided to put pictures scanned by these types of machines on the Web, kind of as a public service. My point is to demonstrate to users of consumer-grade scanners that the major area of potential improvement for generating quality scans lies in increasingdensity range.
As a representative CCD scanner I chose aNikon LS-2000. The drum scans were performed on a Linotype-Hell ChromaGraph S3900. Why did I choose these devices? Mostly by default: an LS-2000 sits on my desk at home; the company where a friend of mine works has a variety of flatbed and drum scanners, but since they have a battery of S3900s, it's easier to get access to one of them.
The Web imposes severe constraints on any author attempting to implement a scanner comparison page. First, when we're speaking of high-end scanner output we're talking big-time files and transmission times. To make retrieval time more tolerable, I've down-sampled and cropped from the full-frame images as much as possible without compromising the comparability of the scans. I've included links to higher resolution picture details for closer comparison. Second, the Web is a mass medium, accessible to users with a wide variety of devices. Most PC monitors are not very capable when it comes to displaying photographic images. Unfortunately there's not much I can do if your video card/monitor cannot display these images. The LS-2000 scans were performed at 8bits/channel and 1200 dpi, well within the limits of its capabilities of 12bits/channel and 2700 dpi. The S3900 scans were performed at 8bits/channel and 1200 dpi, also well within its capabilities.
Comparing the Images
I've included 2 sample images for comparison of CCD and drum scanned images. These images were chosen as a test of the LS-2000's D-Max, one of the highest for CCD scanners. Other than D-Max, these scans are not representative of the ultimate capabilities of either the LS-2000 or S3900 in particular--the image files required for this would be huge--but of CCD and drum scans in general. Technically speaking, I'm showing how images compare when D-Max is the independent variable. Before clicking to display either image make sure your video card/monitor is set to at least 24-bit color.
For each image, compare the CCD and drum scanner images at 1200 dpi to verify that the drum scanner is better at imaging shadow detail at equivalent resolutions. Next, display the same CCD scanner image detail at 2700 dpi. There is a slight improvement in quality over the 1200 dpi image, but it is still not nearly definitive as the drum scanned image. This demonstrates that doubling resolution fails to improve quality sufficiently to compensate for the drum scanner's higher D-Max at 1200 dpi.
Take my word on this, the two drum scanned images closely approximate the slides when viewed on a light table. They would have been dead on if more of the S3900's bit depth and resolution had been used. You might also notice that even at 1200 dpi the S3900 starts to resolve film grain, because of its very high density range and absence of noise.
By the way, did the drum-scanned images look dull and muddy? It's possible your monitor, even though the video card is set to 24-bit color, can't hack showing high quality images created by a drum scanner. That's no insult; it only highlights the essential problem of using drum-scanned images: few consumer level devices can handle the long tonal ranges producible by drum scanners.
Although the drum scanner images contain more shadow detail, that doesn't necessarily make them esthetically more pleasing or more effective at expressing the photographer's vision. Remember the Christmas pictures of the family sitting on Aunt Emma's sofa? Is it really necessary that the sofa's plaid pattern show up in your prints? And if you scan for increased resolution, will your final output device be able to make effective use of it?
The Drum Scanner Option
Where critical scanning and volume are an intrinsic part of business operations--in publishing, engraving, and fine arts--it makes sense to own a drum scanner. For the majority of us who dabble in imaging, a competent CCD desktop scanner will fulfill the vast majority of needs. Personally, I would use a drum scanner for badly exposed shots, larger format film, posters, and archiving of fine art. For these exceptions, it's sufficient to have recourse to a service bureau that performs drum scanning.
The drum scan images, when compared in detail with CCD images, show how good scanner technology could be if more emphasis were to be on improving D-Max. In addition, it makes little sense to do this in isolation without improving monitors and printers in parallel (as you possibly may have seen when you tried to display the pictures). Hopefully this effort will modestly change the way consumers evaluate scanners and ultimately drive the design process of scanners.
I would like to thank Janet Miller of Time Inc. and Brad Pallas of Hachette-Filipacchi Magazines, Inc. for their assistance. Austin Franklin of darkroom.com and Phil Lippincott of AZTEK, Inc. graciously contributed by explaining the distinctions between "density range" and "dynamic range".