Photo Scanner Technology Explained
© 2006 Ken Rockwell. All rights reserved.
INTRODUCTION
I forget that most people coming to this site are new to digital imaging.
I've been studying bits and pixels since 1973, and been working with it daily to earn my living at my real job since the 1980s.
Allow me to explain about bits and pixels and DMaxs so they make sense.
Software alone can't do this as of 2005. Software still fails because software can't differentiate between dirt and a real image element, like a branch, anywhere near as well as our eyes can.
The founder of Applied Science Fiction, now a part of Kodak, had the brilliant idea to use a fourth infrared (IR) scanning channel, in addition to the usual three red, green and blue channels. Color film is transparent to IR, and dirt and scratches are not. BRILLIANT! The scanner knows what looks black to the IR sensor is a film defect.
Locating the dirt is the hard part. Once that's done it's easy to fill in the dirt and scratches automatically with software that patches over the known bad areas.
Because the film's image has to be transparent to IR you cannot use ICE on traditional silver B/W film. Traditional B/W film's silver is as dark to IR as it is to light, thus if you're a wise guy like me and try it you'll get totally soft highlights since the scanner will think all the dark parts of the negative (the highlights) are all scratched and the software will try to patch (smear) all over them. Remember B/W film's image is made with silver, which is why B/W film and prints last forever since metal doesn't fade away. Color film and images just use dye which all fade in time.
Thus ICE works great for color negatives and slides but not B/W. It should work on chromogenic B/W films that process in C-41 color negative chemicals. Ignore me, read your scanner's instruction book and try it. It is not supposed to work on Kodachrome since Kodachrome's dyes are also somewhat opaque to IR. I have not tried it. A reader writes in July 2005 that his Minolta 5400-II works great with Kodachrome; I haven't tried it.
In fact, today the best way to shoot B/W for digital is to shoot it on color film since 1.) ICE works and 2.) you can choose the mixing of the color channels in Photoshop to select the effects that in the old days you had to do before making the image by choosing a colored filter to put over the lens. In the old days I would sometimes shoot several images with different filters; today I just shoot it once on color film if my output is to be B/W digital.
It takes about three times as long to make a scan with this feature turned on. It saves you more time than that since you no longer have to spot the scan.
I like the way ICE works on the Nikon LS-2000 and LS-4000, but on the LS-30 it seems to dull the image. People have differing opinions on this feature.
I wouldn't buy a scanner for 35mm unless it has this feature.
It does not soften the image, but it may take some little one-pixel bites out of sharp lines. I usually use ICE, although if the image has no broad areas where dirt would be obvious I may turn it off to save scan time.
Canon's FARE system on the 2400 flatbed is awful. It replaces dirt with big blobs and erases other parts of the image. I found it to be useless except in broad sky areas, where Photoshop's dust filter also works perfectly. Use FARE on a slide with any detail in it and you will see all sorts of screwy artifacts once you learn what they look like.
Light Sources (LED vs. Cold Cathode Fluorescent)
Don't worry. The scanner makers worry about this so you don't have to. A scanner's performance depends on many, many, many factors and the interaction among many more factors. The key determinant in a scanner's performance is the designer's ability to balance these many factors to give the best performance at any price level. Yes, even $100,000 scanners have to make engineering compromises, thus any scanner's worth depends on how clever the engineers were at making good choices.
Don't be distracted by what kind of light bulb it uses or what brand of lens or any of that.The only way to determine the scanning ability of a scanner is to make some scans and look at the images.
These are all very different terms.
Resolution are the finest details between which a scanner can see. In other words, the scanner needs to be able to resolve the difference between very close together image elements.
Some film scanners are so sharp they can see details smaller than the size of a pixel. This is called "aliasing" or "false resolution." It can lead to emphasizing grain from print film. It also allows scanners to see details finer than their image structure (defined below). This is beyond what I feel like explaining here, but suffice it to say that even though one can see details finer than the pixel structure that they are misrepresented in the image and no longer look a fine as they do on the original. They look like Moire patterns if you are looking at test charts, or extra grain if you are scanning negatives. This usually only happens at the lower resolution settings of a scanner if the designers decided not to go to the efforts of doing the proper Nyquist anti-aliasing.
Most people and manufacturers erroneously use the term "resolution" when they really are referring to the image structure, or the number of pixels in an image defined below.
For instance, DPI or dots per inch, referred originally to the the ability of phototypesetting image setters to write small details and smooth looking type. Today this phrase refers to how many pixels one has, which is image structure (defined below) and not resolution.
Definition is a subjective term referring to how sharp something looks.
Image Structure refers to how many pixels an image has. All because an image has a certain number of pixels does not mean that all these pixels are different and can differentiate detail, or resolve parts of the image. For instance, it is trivial for scanners and cameras to add meaningless pixels as explained elsewhere. This adds to the number of pixels but does not add anything to the resolution or definition.
"Real" pixels are real pixels.
Phony pixels are made in several ways. Many scanners and digital cameras add pretend pixels in between the real pixels to help make an image smooth and not pixely when enlarged, however they do not add any sharpness or detail.
All scanners can add phony pixels when set to resolutions above their legitimate optical resolution. Digital cameras do this when set to larger (but bogus) "recording resolutions" or "interpolated resolutions." You can do this yourself in a program like Photoshop when you resize the image and have "resample" box checked.
For scanners and digital cameras these claims are simply lies on the parts of the manufacturer and they are rampant today. Altimira's Genuine Fractals does the same thing: it makes pretend pixels and eliminates the obvious boxy pixely look, but does not add any more detail to your image as you enlarge.
Make very sure you are trying to compare legitimate resolution to legitimate resolution among scanners. Always ignore the larger of two numbers, for instance, a 2400 x 1200 DPI scanner is just a 1200 DPI scanner, and just forget about any "Interpolated" resolutions. Those are meaningless.
Dmax is the maximum absolute density of a piece of film. Velvia goes to almost 4.0D in its blackest blacks, the deepest black in any color film. Most slide film only goes to about 3.5D.
Color and B/W negative films only go to about 1.4D perfect negatives, and in even heavily exposed ones almost never over 2.0D, so DMax has no importance for scanning from negatives.
D range is the range between light and dark. Since 4x5" and 120 transparency film usually has a Dmin (clear area) of about 0.05D then a scanner with a Dmax of 4.0 really only would have a D range of 3.95.
As you can see, it looks better to spec Dmax and not D range. Dmax is the hard one to get. Only an idiot designs a scanner not to be able to scan to 0.01 D min.
D range is a more meaningful spec than D max, and here's why: Remember that marketing departments may choose to measure Dmax with the analog gain or the light bulb turned all the way up, in which case you can see into deeper blacks, but will lose the highlights. You have to read and ask carefully, everyone lies differently. If you see a spec of Dmax 4.2 and D range 3.9 that means that the D max of 4.2 is really a hoax. It means you really can't get to D max of 4.2 unless your highlights are dull gray at 0.3D. The only way a scanner like that gets to see 4.2D is by turning up the light bulb a stop and blowing out highlights at less than 0.3 D. With a decent transparency this hypothetical scanner really only has an effective D max of 3.95, with a D min of the good transparency probably 0.05D.
Here's more reasons to ignore a D max spec: What does the scanner do at that density? Does it have a 20dB signal-to-noise-ratio, or is that where the noise completely covers the data from the film? Is that the quiescent noise level of the scanner, or can one still see clearly what's on the film? Since no one in CCD land specs the criteria on how they determine their Dmax spec they are all meaningless.
Grasp as we may for meaning from specifications, the only way to know how well they can be made to scan is to try for yourself with a dense example of a Velvia transparency. For instance, I tried both a $300 Epson 1640SU photo (3.0 Dmax spec) and a $1,500 Microtek Artix 1100 (3.9 Dmax spec.) The two were about the same, and in fact, the noise from the Microtek was nastier because it caused visible streaks. Microtek claimed that that scanner was defective, and I have not tried other samples. I tried the same dark night Velvia transparency on those two scanners, as well as a Nikon LS-2000 and Coolscan III LS-30 and Kodak Photo CD. They were all subtly different, but nowhere near as different as the D max specs of these scanners would lead you to believe. Maybe if you are all very good (and you have been) and I get very bored I'll just post the results up here for you all to see.
Scan Times
Don't believe what the manufacturer says. Most scanners will take several minutes to scan whatever it is you want to scan. Manufacturers lie about the scan times by specifying the times at a lower resolution and for a smaller image then you will be scanning, and then lie about that, too.
You can safely forget about how many bits (8, 10, 12, 14 or 16 per channel, or 24, 30, 36, 42 or 48 bits per sample) a scanner claims to have. These have nothing to do with how good your scan will look.
The illustrations on the boxes of some Microtek scanners and sales literature for theMinolta Multi Pro are complete lies, plain and simple.
Color depth has nothing to do with shadow detail or color accuracy.
Yes, you can see more shadow detail in the illustration at the Minolta link, but that's because Minolta is deliberately deceiving you by having played with the same image in Photoshop to lighten one! In court Minolta would plea that those images were for illustration only and that they were manipulated only to allow the differences to be more visible and were not intended to deceive. Sure.
If you prefer to believe a salesman or your hobbyist friend who is an engineer in some unrelated field, fine, but if you want to listen to someone who does this for a living or can believe your own eyes read on.
Even the dumpiest scanners today brag about 42 bit scanning. That means that they use an analog-to-digital converter that just happens to have 14 bits coming out of it to digitize each of the signals from the analog outputs of the CCD preamplifier for each of three color channels (red, green and blue.)
If this sounds technical, it is. If this makes no sense to you, that's my point. I used to work where we made the A/D converters for all the world's best scanners and I can, and probably will, explain this in excruciating detail someday. What is important for you to know is that there are so many factors that relate to scanner quality that the number of these internal bits alone is completely insignificant. Skip to the end of this section for the real answer if this gets too technical.
Last years models were 36 bits (12 bits per channel) and the year before that it was 24 bits (8 bits per channel).
Let's explain why more bits are better, then I'll explain why this make no difference on scanners in the under $3,000 price range.
We only can see between 6 to 8 bits per channel log (18 to 24 bits per pixel) at best.
The reasons a scanner would like to have more than 8 bits per channel is that is so that one can 1.) convert digitally from linear analog input to log digital output instead of doing the process in analog (in other words, resolve fine differences in the black areas of the image) and 2.) to allow further fine-tuning of an image in the higher resolution so that one still has 8 good bits per channel (24 bits per pixel) after all the image tweaking is completed.
One looses accuracy in every 8 bit Photoshop operation, even though all the bits are still there. Mathematical errors accumulate with every operation, just as you would if you always had to write down your results from balancing your checkbook with only 4 digits. You need 16 bits to represent the product of two 8 bit numbers, so when the answer has to be truncated or rounded or redithered back to to 8 bits you loose something. If you only start out with 24 bits, by the time you get done in Photoshop you have lost a few bits of accuracy due to all the small mathematical errors. By starting with and working with more buts you can easily preserve your accuracy. This becomes critical if you are playing with levels and curves.
Scanning in the 42 bit mode into Photoshop lets you attempt to retain plenty of bits of accuracy even after a lot of Photoshop tweaks. This is good.
Here's the problem: The noise of the CCD and it's circuitry completely cover up any of the precision the A/D converter might have. At best the noise level of a typical scanner is at about 30 bits per pixel, and probably even worse. ALL THE EXTRA BITS ARE DOING IS REPRESENTING NOISE.
If these were $50,000 photo-multiplier tube (PMT) based scanners you might have enough signal-to-noise ratio coming from the image sensor (the tube) to justify a 12- or 14-bit per channel A/D converter
These CCDs simply don't yet have the signal to noise ratio to justify the high-bit ADCs. That's OK, since the ADCs are not likely to be accurate to as many bits as they claim as well.
Therefore, ignore people who work in computer stores and try to get you to believe that more bits are better, and can't even explain exactly what the extra bits do. In this case they mean nothing.
Even worse, unless you are using a pretty hot version of Photoshop, very few computer programs even can read anything other than 24-bit (8 bit per channel) images. Unless you have Photoshop 6.0 you can't do much of anything with 42 bit images, even if you can read them.
The JPEG (.Jpg) format only works in 8 bit mode, for example. If this is your case, the only advantage to more real bits inside a scanner is if the scanner was doing some signal processing in the digital domain in which case more bits could be better.
Hopefully by now I've convinced you that these internal details are meaningless taken outside the context of the entire system design of a scanner. All things being equal more bits are better, however things are so unequal that the number of bits is meaningless.
What is this Drum Scanner against which all other scanners compare themselves? (so you know they're not as good)
Professionals in print shops that produce all the magazines and printed catalogs we read laugh at the CCD-based scanners we photographers use. All the scanners you or I are likely to buy are based on CCDs, the same little chips that you have in your camcorder or digital camera. The only difference is the scanner CCD is much longer than the one in your camcorder (more pixels) and it's only one line that scans across your image slowly to make a complete scan.
Drum scanners spin your film around on a drum while a fixed laser or other beam of light looks at the art as it spins. The beam is them picked up by a very sensitive vacuum tube called a Photo Multiplier Tube (PMT). This big single, fixed tube is much more sensitive to light than any of the teeny-tiny pixels on a CCD and therefore can see a broader range of light from white to black, and also sees it without the noise of a CCD.
Drum scanners are good not because of the drum, but because the image is picked up by a much more sensitive PMT.
The big PMT is a zillion times more sensitive to light (read shadows in slides) than a moving CCD with teeny weeny pixels. This gives far cleaner images right into the blacks of Velvia. The microscopic pixels of the CCD fight for each photon. There are so few that honestly you start seeing the variations of single photons as noise. There are bazillions of photons that hit the huge PMT from the single bright beam that blasts the original, so low noise and cleaner scans are the result.
See Hamamatsu's site for more information here.
A drum scanner beam can be focused as small as you want. You can move the drum really slowly and get almost unlimited resolution.
Drum scanners used to cost hundreds of thousands of dollars, were about twenty feet long and took years to learn to operate.
Today new ones from Aztek start at about $20,000. Heidelberg still makes the $100,000 ones, unfortunately their website is sucky so good luck finding anything. Also look at ICG and Screen. I think you can get used from about $5,000, but beware, I hear these things are a royal pain to operate.
Of course you have to soak the film in oil before mounting the film to the drum, and then clean the film afterwards.
These things run on vacuum tubes so probably always need to be screwed with.
If regular scanners confuse you enough with color profiles and pixel densities, then don't even think about a drum scanner because you have even more new ways to screw up your scans, like selecting the correct scan apertures.
If you want it scanned as well as possible, drums are the way to go.
The Imacon, advertised as a drum scanner because they use a curved film path, is nota drum scanner. It's just marketing people trying to mislead you again.
You send your film out for drum scanning. I intend to try using photographer Mike Strickler in Northern California. Look under "Services" and then under "Digital Services" and then "High Resolution Drum Scanning." Likewise you can try local publishing and printing houses, but be very careful. Print shops can muck up and scratch your film since they're just doing this for commercial printing day in and day out and won't be going out of their way to treat your prized film as you would. I'm hoping a photographer like Mike takes the time to treat our film as carefully as his own.
How to Scan
Dynamic Range Trick for Negatives
If the dynamic range of a negative results in washed out highlights or dead shadows with the scanner set to scan a negative, trick the scanner and set it to positive (or transparency) mode for the scan and then invert it in Photoshop. After this, go into LEVELS in Photoshop and set your highlights and shadows where you want them.
No comments:
Post a Comment