Choosing a Scanner
At this point in time (November 2002) the landscape for film scanners is vast: anything goes, from CCD flatbeds to multi-hundred-thousand dollar PMT drum scanners. While it may be true that money can (and does) buy quality, there are often exceptions to this rule (though don't expect a $400 flatbed scanner to rival a Tango drum scanner most of the time).
How Much to Scan?
An often-asked question is how large should one scan. That is, what resolution and how large should the file be? And the answer to that lies in the ultimate use for the file.
The prevailing wisdom these days is to scan at the highest resolution possible and perform all dust spotting and color correction on this file. Once the basics are covered, archive this hi-res file and work on copies for the output that you require, downsampling as necessary. Since scanning and dust spotting/color correcting takes quite a bit of time it makes the most sense to do it once rather than making extra work every time a print or web image is needed.
The downside to this is that the files can be huge. A 16 bit/pixel, 4000ppi scan is somewhere around 120MB. Not trivial. But hard drives and blank CDs are cheap. How much is your time worth?
There are several ways to view the scanner world:
Now, the goal of scanning is always the same: digitize an analog original so it can:
Film scanners are dictionary-sized contraptions that connect to a computer via FireWire, USB, SCSI, or Parallel interfaces. Often this is the best route for most home users as it affords the greatest quality scan without having to jump to the stratosphere of drum scanners. Most film scanners work more or less the same way with a slot in the front of the box that accepts mounted slides or strips of film. The scanning array is a line of 3-color CCDs one pixel wide that move across the film by stepper motor. Sometimes it is the film holder or film strip that moves instead of the CCD array. But either way, the result is the same, the film passes over the imaging sensors and the data is sent to the computer.
Because film scanners scan the original image (the film) via light passing through the image the sensors are able to capture a much wider dynamic range (on the scale of 1000:1 or more) than scans from a print. The end result is that highlights aren't blown and shadows retain detail.
Flatbed scanners are what most people have encountered in their lives at one point. The local Super-Mega-Department-Store often sells them for $59.99 (with a $20 rebate). Flatbed scanners are useful for scanning reflective materials like prints and flat artwork (and magazine pages for those that like to maintain homage pages of their favorite star(let)).
The problem with flatbed scanners comes from the fact that often the user is scanning reflective material, which only has a dynamic range of 300:1 or less. Therefore, you don't get the pop that you would get from scanning the original film. It is possible to get flatbed scanners that have illuminated lids with which you can scan transparent originals (negatives or slides). Some are very good within their respective limits (Epson's 2450 or 3200 comes to mind). But since you will always be scanning through a piece of glass there is always (or often) going to be a reduction in quality over using a dedicated film scanner. Especially with 35mm film where the original isn't that large to begin with. But for economy scanning of medium format film (or just economy scanning of prints) the flatbed is often the least expensive route.
For those that run a business. Or have way more money than brains there are PMT drum scanners. Drum scanners operate by placing the original artwork on a clear drum via wet mounting. The wet mounting works by sandwiching the original between the clear drum and a piece of acetate using mounting oil. The drum is then placed in the scanner and spun at rotations up to 500-1000 rpm. Light passes though the original (if it's film) and the photo-collector reads the photons as they hit the sensor, multiplying the intensity to capture the full range of density information on the film. For reflective materials the process is similar but the sensor is on the outside of the the drum rather than on the inside of the drum (these scanners have two sensor locations).
Drum scanners have true resolutions of 11,000ppi+ and are able to capture detail well beyond the range of the naked eye. Because of the nature of drum scanners, the output is often far and away the best of all of the technologies. But given that most new drum scanners cost US$100,000+ they are well out of the price range of most users for individual use.
The specs that manufacturers often cite in their glossy brochures are (as always) to be taken with a grain of salt. The important numbers are the two usual suspects that everybody cites: optical resolution and dynamic range.
Optical resolution is the easiest of the two to dissect. Most flatbed scanner manufacturers list resolutions for the X and Y dimensions that aren't the same while most film scanners have a single resolution listed. For flatbeds, the manufacturers are trying to talk up their scanners by listing interpolated resolution. The scanner often only has the same optical resolution in both dimensions but the specs will list something like 600dpi x 1200dpi. In fact, the true resolution is only 600ppi x 600ppi with the extra 600ppi coming from either the manufacturer's software or the manufacturer playing tricks with half-stepping of the stepper motor in the scanner itself. This more or less creates pixels out of thin air, giving no more real information. When you see this kind of spec for the scanner's resolution, the lower number is always the true optical resolution and this is what matters (I'm not aware of any cases where this isn't true).
For film scanners, the manufacturers almost always cite the true optical resolution. Often 2700ppi up to 4000ppi (or higher). In this case, the manufacturers aren't lying to you. Film scanners are capable of this resolution. I would attribute this lack of deception on the fact that those that buy film scanners often pay more so they more thoroughly research their purchases beforehand, thereby making it an almost useless act for a manufacturer to get away with questionable marketing. But as always: caveat emptor.
The Math of Dynamic Range
As an aside, for those interested in the math, the dynamic range numbers work out as such:
Dynamic range is based upon a log(10) scale while A/D converters are in binary, so based upon a log(2) scale. The dynamic range comes out of the 2^(bit number). So for a 10 bit A/D converter, the scanner can have a dynamic range of log(10) of 2^10 = log(10) of 1024 which is 3.0103. A 14 bit A/D converter is 2^14 = 16384. log(10) of 16384 is 4.2144.
The other biggie spec is dynamic range (and Dmax) and here's where the waters are very, very muddy. Dynamic range simply refers to the range of lightness between the brightest highlight (white) and the darkest shadow (black). Most slide film can contain a range of 3.2. Negative film, on the other hand compresses a wide dynamic range into a much narrower band, which means a scene that might have had intensely bright highlights and very dark shadows now evens them out to a flatter range of approximately 2.8. The thing to keep in mind is that it's much easier for a scanner to scan from film with a smaller dynamic range than a larger one which is why negative film often poses less of a problem to film scanners than slide (reversal) film.
The problem with dynamic range specs is that there is no standard for measurement so often a manufacturer will list the theoretical dynamic range for the scanner's analog to digital (A/D) converter. The problem with this is both obvious and sublime. The obvious issue is that theory is never (or almost never) what happens in real life. The sublime issue is that A/D converter is almost never the limiting factor but instead the scanner's sensor. So when a manufacturer like Nikon lists a dynamic range of 4.2, they're listing the theoretical range of their 14-bit A/D converter and the reality is that the spec is more likely to be in the mid-3.x range due to limitations of the sensor. The only way to know if a scanner can do what you want (in terms of dynamic range) is to scan your most horridly difficult light-to-dark slide and see the results for yourself.
For more information check out Wayne Fulton's amazingly succinct and thorough description of dynamic range at his Scantips site.
ICE³ from Applied Science Fiction and FARE from Canon are pretty much the same thing (in at least one feature): dust removal via hardware. In the dust removal, the film is scanned by both visible light sources and via an infrared means. This allows the visible light to see everything, including color information and dust and the infrared sees the dust. This information is then compared and interpolated, allowing software to see the difference and 'heal' the areas that have dust. As a result, specks are virtually (or in reality) removed from the scan, saving a boatload of time in post-processing. The downfall of this feature is that the 'healing' process can (and does) result in softening of the image. Often, Canon's FARE affects the image less than Nikon's/ASF's ICE³. What's more, dust removal also adds time for scanning as the software processes the information before it writes the image data to a file. This can add a significant amount of time (up to 50% longer scan times). So keep this in mind when thinking about these features. In a lot of cases this additional scanning time is nothing compared to spotting the print with the clone tool in Photoshop not to mention the wear and tear on your wrist as you blot out those pesky flaws.
One difference between FARE and ICE³ is that ICE³ also has provisions for grain reduction and a restoration of image colors (called GEM and ROC respectively). These features do pretty much what their names indicate but with the added issue of continuing to reduce the 'sharpness' of the un-touched image. And just like dust removal, these features add yet more time to the scan.
SilverFast from LaserSoft Imaging is a robust software package for scanning. SilverFast has a rather large list of supported scanners and provides a strong but not overly-confusing graphical interface. SilverFast also has strong support for ICM/ICC profiles, allowing for improved color managed workflows.
The underdog of third-party scanning software, VueScan is a often cited by both amateur and professional photographers as the package to get once you've realized that the software that came bundled with your scanner isn't all that and a bag of chips. The user interface for VueScan is probably more dense than either the manufacturer's software or packages like SilverFast but it still allows for a lot of adjustment as needed. But the learning curve can be steep. The plus side is that VueScan often supports features that the manufacturer's software does not, like single or multi-pass multi-sampling.
VueScan is a standalone application that has no Photoshop plug-in to allow direct scanning from Photoshop. But often this is preferred for batch scanning workflows so to many this isn't a hinderance.
Well, it's free. And supported by the manufacturer. But not always the best option for some. Often the software bundled with the scanner doesn't support all features of the hardware (the Minolta Dimage Scan scanners with Minolta's software come to mind). But the user interface if often less daunting than VueScan. Further, it is rate to find built-in color management capabilities in the manufacturer's software (NikonScan is at least one of the exceptions). And any problems that may arise can be covered by customer support.
Often, manufacturer's software is only a plug-in for Photoshop but in some cases, the package can also be run as a standalone application.
The sub-$500 range holds a lot of flatbed scanners and a couple of decent film scanners. If you know you won't need to scan flat artwork, magazines (ahem), or prints then it would probably be prudent to stick with a film scanner. The Minolta Dimage Scan Dual III might be the best option with 2820ppi and USB (v1.1) interface it has a decently capable dynamic range and good scanner software. In addition, the Canon FS 2710S and FS 2720U can still be found for less than $500 from reputable online retailers like B&H Photo and Adorama. Keep in mind that the FS 2720U will not work under Mac OS X even with VueScan.
If you think you'll need a flatbed for prints or if you want to scan medium format film in a low-cost manner then the Epson Perfection 2450/3200 Photo is far and away the best choice in this price range. The scanner has a 2400/3200ppi resolution and dynamic range that is far better than should be allowed by law(s of physics and economics), this scanner will allow you to make prints up to 12"x12" from a 6x6 negative.
In this price range the contenders start to get serious. Canon's FS4000US is often cited as the best in this range with its 4000ppi resolution and built-in dust and scratch removal feature (called FARE). Other scanners from Minolta (the Dimage Scan Elite II) and Nikon (the Coolscan IV) are in the 2800-2900ppi range and have the potential wide dynamic range and high-bit editing with 14 bit or 16 bit A/D converters. In addition the Nikon has ICE3 dust removal for those that have a lot of film to scan and little time to remove the defects via the clone tool.
In this price range there really is pretty much only one scanner: the Nikon Super Coolscan 4000ED. This scanner has a 4000ppi resolution and a very strong ability to pull details from shadows. In addition, this scanner can take advantage of single-pass multi-sampling to reduce CCD noise in the shadows of a given scene. Like the Coolscan IV, this scanner also has ICE3 for infra-red dust and scratch removal. The biggest downfall of this scanner is probably it's terribly narrow depth of field that can cause even slightly warped films to go out of focus in parts of the frame. For this reason, a number of people prefer Canon's FS4000US.
Also in this price range is Polaroid's SprintScan 4000+ and Micortek's ArtixScan 4000t. These machines are the same beast but with different badges. In reality the SprintScan line was developed by Polaroid and then the manufacturing rights were sold to Microtek who now manufactures all the machines, including the Polaroid-branded scanners. Regardless, this scanner is very capable with a very, very low-noise CCD imager. Multi-sampling is possible with this scanner but it is rarely needed as the imaging chip is that good. Another perk of this scanner is the fact that it is sold with SilverFast, a scanning software package that allows a huge amount of control with a strong user interface. Another added bonus is that this scanner is often a few hundred U.S. dollars less than the Nikon LS-4000.
In this area the heady world of high-end CCD scanners and PMT drums opens itself. Often this range is for pre-press businesses and people with more money than brains. Among the scanners are Heidelberg's line of PMT drums and Imacon's CCD-based pseudo-drums. In addition, CreoScitex has a line of very good flatbeds for those that need to scan both flat artwork as well as film in high resolution but perhaps don't want to pay the price (or can't afford) a drum scanner.