|[ Team LiB ]|
Beside their basic mechanism, scanners are distinguished by their features. Among these are whether the scanner can produce color images, its scanning speed, the dynamic range it can handle, its resolution, and whether it can recognize text characters and translate them into character rather image data. The availability of options such as transparency adapters, sheet feeders, and optical character recognition software also makes one model more suited to some applications than others.
Early color scanners give monochrome models a hefty edge in performance. The earliest color scanners were three-pass machines. That is, they required three passes to make a complete image, one pass for each of the primary colors. These ancient scanners used three separate light sources of different colors and took a full scan in each color. Nearly all modern scanners use one-pass designs. They have a single light source and rely on filtering in their photodetectors to sort out the colors. One-pass color scanners can operate just as quickly as monochrome models, although transferring a large color image measuring dozens of megabytes still takes longer than moving a monochrome image one-third its size.
The speed at which the scanning CCD moves across the image area is only one factor in the total time required to make a scan. Most scans require at least two separate passes. First, you command the scanner to make a pre-scan, which is a relatively quick, low-resolution pass across the image that helps you establish its brightness range and also lets you target a specific area for scanning. Then you make the actual scan at the resolution you want.
In addition, the interface used by a scanner influences the speed of scans, as noted later in the chapter. The high-resolution bit-images produced by the scanner represent a huge amount of data—megabytes—and a slow interface constricts this data's flow.
If the scan you want to make is large, you also have to wait for image processing, both in the scanning software and in the host application. Very large scans can add minutes or more to the total scan time if you exceed the memory capabilities of your computer. Although Windows can take advantage of virtual memory to let you capture images of nearly any size, this technology uses your disk drive for extra storage space, which adds the seeking, writing, and reading times to the total time of your scan. If you plan to regularly make large scans, you'll speed things up more by adding memory to your computer—in the dozens of megabytes—rather than looking for a faster scanner.
At one time, the base-level distinction between scanners was like that of television sets—color or monochrome. And, as with televisions, the monochrome variety is almost extinct. But color scanners aren't all equally colorful. Some recognize more hues than others.
The compass of colors a scanner can discern is termed the scanner's dynamic range. The most common means of expressing dynamic range is bit-depth, the number of bits needed to digitally encode the total color capacity. Most common scanners can distinguish 256 (8-bit), 1024 (10-bit), or 4096 (12-bit) brightness levels in each primary color. Just to make their scanners seem more capable, scanner manufacturers add up the total number of colors within the scanner repertory, so you'll see 24-bit, 30-bit, and 36-bit color scanners.
The actual dynamic range and the bit-depth of a scanner are not necessarily the same. A high-quality scanner will be able to resolve the number of brightness levels its bit-depth implies. The bit-depth actually specifies the range of the analog-to-digital converters that convert the level detected by the scanner's CCD sensors into digital signals. Taking advantage of that bit-depth requires that the scanned image be properly focused on the CCD sensor under optimal illumination. If the focus of the scanner's optics is off, pixels will blur into one another, which lowers image contrast and the dynamic range of the scanner. Similarly, if the illumination provided for the image during the scan is uneven, the variations will wipe out some of the available brightness levels of the dynamic range. Consequently, two scanners with the same number of bits quoted for their dynamic range may, in fact, have different actual dynamic ranges.
Most computers can, of course, display from 256 to 16.7 million different hues (that is, 8-bit to 24-bit color). When that is more than you or your software wants to manage, the scanner's palette can easily be scaled back, either through hardware controls or through software, to an even smaller bit-depth. With even a minimal 24-bit scanner capable of giving your software more than enough color, the extra bits of higher-cost scanners might seem superfluous.
Those extra bits are very useful, however, when the scanner preprocesses image data before passing it along to your computer. A scanner with 36-bit dynamic range can capture all the shades and hues of an image and let you process them down into 24-bit color for your computer to use. You get to choose how to handle the conversion, compressing the dynamic range or cutting off colors you don't want to use. The extra bits ensure that your scanner can capture all the detail in the darkest shadows and brightest highlights. When you use a transparency adapter, greater dynamic range helps you compensate for thinner or denser originals, potentially yielding workable scans from transparencies you can barely see through.
The trend among high-end scanners is toward a dynamic range of 48 bits, giving you a greater range of manipulation. Note, however, that most existing software for image manipulation cannot yet handle 48-bit images (although the current release of Adobe Photoshop can).
Many scanners have automatic modes through which they determine the proper brightness and contrast ratios to take best advantage of the translation of the scanner's dynamic range into the 24-bit level of color (or other level of color) used by your computer. The most common means of making this optimization is to pre-scan the image. The scanner then checks the brightest and darkest points of the scanned area. Using these values to establish the actual range of brightness and color in the image, the scanner can adjust its transformation to yield the image with the greatest tonal range to your applications.
With slide scanners, another factor influences your ability to work with marginal slides and negatives—the maximum image density, usually abbreviated as D-max, that the scanner can handle. A more dense image is darker—there's more dye or silver in the image. Because slide scanners work by shooting light through the slide or negative, a very dense image may prevent any light from getting through at all. The D-max indicates how dense an image can be before the scanner can no longer distinguish the light shining through it. Any part of an image that's more dense than the scanner's D-max rating blocks up as solid black (or, if you're scanning a negative, solid white).
Technically speaking, the density of a photographic negative or slide is the ratio of the intensity of light shining through the image over the intensity of light that actually gets through, expressed as a logarithm. The scientific formula for density is as follows:
As a practical matter, the best of today's slide scanners cope with a D-max of about 4.2. That means they can detect light that's diminished by a factor of more than 10,000. A scanner with a D-max of less than 3 will have difficulty dealing with the full range of image brightnesses, even on properly exposed film.
Scanners differ in the resolution at which they can capture images. All scanners have a maximum mechanical limit on their resolution. It's equal to the smallest step that their sensor can be advanced; typically a minimal scanner will start with about 300 dots per inch and go up from there in regular steps such as 600, 1200, then 2400 dots per inch. Special-purpose slide scanners achieve resolutions as high as 10,000 dots per inch. Because it represents the limit of the quality the scanner hardware is able to resolve, this measurement is often termed the hardware resolution of the scanner. Another term for the same value is optical resolution.
Beyond the mechanical resolution of a given scanner, the control software accompanying the scanner often pushes the claimed resolution even higher, to 4800 or even 9600 dots per inch, even for an inexpensive scanner. To achieve these higher-resolution figures, the control software interpolates dots. That is, the software computes additional dots in between those that are actually scanned.
This artificial enhancement results in a higher resolution value quoted for some printers as interpolated resolution. Although interpolating higher resolution adds no more information to a scan—which means it cannot add to the detail—it can make the scan look more pleasing. The greater number of dots reduces the jaggedness or stair-stepping in the scan and makes lines look smoother.
The new dots created by interpolation add to the size of the resulting scanned file, possibly making a large file cumbersome indeed. In that interpolation adds no new information, it need not be done at the time of scanning. You can store a file made at the mechanical resolution limit of your scanner, then later increase its apparent resolution through interpolation without wasting disk space storing imaginary dots.
As with colors and shades of gray, a scanner can easily be programmed to produce resolution lower than its maximum. Lower resolution is useful to minimize file size, to match your output device, or simply to make the scanned image fit on a single screen for convenient viewing. Although early scanners and their control software shifted their resolution in distinct increments—75, 150, and 300 dpi, for example—modern scanner-plus-software combinations make resolution continuously variable within wide limits.
The actual hardware resolution of a scanner is fixed across the width of the image by the number of elements in the CCD sensor that determines the brightness of each pixel. The hardware resolution along the length of the scan is determined by the number of steps the CCD sensor takes as it traverses the image area. The size of these steps is also usually fixed. Scanning software determines lower as well as higher resolution values by interpolating from the hardware scan. Consequently, even when set for 50 dpi, a scanner will sense at its hardware resolution level, deriving the lower-resolution figure through software from its higher capabilities.
As with people, scanners are not blessed with the ability to see in the dark. To make a proper scan—that is, one that doesn't resemble a solar eclipse in a coal mine—the scanner needs a light source. All scanners have their own built-in and usually calibrated light sources. In drum and flatbed scanners, the light sources are inside the mechanism, typically one or three cold cathode tubes that glow brightly. Handheld scanners often use light emitting diodes (LEDs) as their illumination source. In any case, in normal operation the light reflects from the material being scanned, and the CCD sensors in the scanner measure the brightness of the reflected light.
Some source materials fail to reveal their full splendor under reflected light. The most important of these are transparencies such as photographic slides or presentation foils. These are designed to have light shine through them (that is, transmitted light).
To properly scan these media, the scanner must put the media between its light source and its sensor. Slide scanners have the source for transmitted light built in. Most other desktop scanners have an optional secondary source for transmitted light called a transparency adapter. The secondary light source tracks the CCD sensor as it scans across the image, but from the opposite side of the original.
Most commonly the transparency adapter takes the form of a thicker cover over the glass stage on which you lay your originals to be scanned. A few scanners have add-on arms that scan over the top of the transparencies you lay on the stage. The latter style works well but does not hold original transparencies as flat against the stage as do the former.
Optical Character Recognition
Scanners don't care what you point them at. They will capture anything with adequate contrast, drawing or text. However, text captured by a scanner will be in bit-image form, which makes it useless to word processors, which use ASCII code. You can translate text in graphic form into ASCII codes in two ways—by typing everything into your word processor or by Optical Character Recognition (OCR). Add character-recognition software to your scanner, and you can quickly convert almost anything you can read on your screen into word processor, database, or spreadsheet files. Once the realm of mainframe computers and special hardware costing tens of thousands of dollars, OCR is now within the reach of most computers and budgets.
Early OCR software used a technique called matrix matching. The computer would compare small parts of each bit-image it scanned to bit-patterns it had stored in a library to find what character was the most similar to the bit-pattern scanned. For example, a letter A would be recognized as a pointed tower 40 bits high with a 20-bit wide crossbar.
Matrix matching suffers a severe handicap—it must be tuned to the particular typeface and type size you scan. For example, an italic letter A has a completely different pattern signature from a roman letter A, even within the same size and type family. Consequently, a matrix-matching OCR system must have either an enormous library of bit-patterns (requiring a time-consuming search for each match) or the system must be limited to matching a few typestyles and fonts. Even then, you will probably have to tell the character-recognition system what typeface you want to read so it can select the correct pattern library. Worse, most matrix-matching systems depend on regular spacing between characters to determine the size and shape of the character matrix, so these systems work only with monospaced printing, such as that generated by a typewriter.
Most of today's OCR systems use feature matching. Feature-matching systems don't just look and compare; they also analyze each bit-pattern that's scanned. When it sees the letter A, it derives the essential features of the character from the pattern of bits—an up-slope, a peak, and a down-slope with a horizontal bar across. In that every letter A has the same characteristic features—if they didn't your eyes couldn't recognize each one as an "A," either—the feature matching system doesn't need an elaborate library of bit-patterns to match nearly any font and type size. In fact, feature-matching recognition software doesn't need to know the size or font of the characters it is to recognize beforehand. Even typeset text with variable character spacing is no problem. Feature-matching software can thus race through a scan very quickly while making few errors.
The typical OCR application involves transferring the information content of multiple pages into electronic form. You must, of course, scan each page separately to derive its information content. With long documents, the chore is time consuming and usually not the most productive way to spend your working hours.
A sheet feeder automatically runs each sheet of a multiple-page document through a scanner. Although a sheet feeder is easiest to implement with a drum scanner, because the scanner has to put the paper in motion anyway, some flatbed scanners have built-in or optional sheet feeders as well.
Sheet feeders are useful primarily for OCR applications. Graphic scanning usually involves individual setup of each page or image. Actually loading a page into the scanner is a trivial part of the graphic scan. Adding a sheet feeder to a scanner used primarily for graphics is consequently not cost effective.
Sheet feeders require loose sheets. They cannot riffle through the pages of a book or other bound document. When a job requires high productivity and the information is more valuable than the printed original, some people cut apart books and similar materials for scanning using a sheet feeder. In any case, you'll probably find a staple-puller to be a worthy accessory to your sheet feeder.
At least six different interfaces designs are or have been used by scanners: Small Computer System Interface (SCSI), General-Purpose Interface Bus (GPIB), standard serial, parallel, USB, and proprietary. Almost all current products rely on parallel, SCSI, or USB connections.
Parallel models plug into legacy printer ports. Most have special cables or connectors that allow you to link your printer to the same port used by the scanner. With modern port/driver software, parallel-interfaced scanners are the easiest to get running—they come to life almost as soon as you plug them in (and install the drivers). The parallel interface is also inexpensive.
The downside of the parallel connection is performance. It is the slowest of the scanner links, and it may double the scan time of a typical page.
The SCSI interface, on the other hand, is fast—the fastest scanner connection in use. The penalty is, of course, the need to tangle with a SCSI connection. This need not be a problem. When the SCSI-based scanner is the only device plugged into a SCSI port, getting the scanner to work is about as easy as with a parallel port. Adding a scanner to a long SCSI chain is as fraught with problems as linking any additional SCSI device.
The other problem with SCSI-based scanners is that they require a SCSI port. Most SCSI-based scanners come with their own SCSI host adapters and cables. (This is one reason the SCSI interface adds to the cost of a scanner.) Installing the adapter in an expansion slot complicates the installation process. Worse, if you want to use a SCSI scanner with a notebook computer, you'll need to purchase a PC Card SCSI adapter.
USB scanners fit in the middle. Although they require a free USB port, nearly all new computers (including notebooks) have at least one. Most have at least two. Scanners fit readily into USB's Plug-and-Play system, but they suffer from the same teething difficulties as other USB products. Although USB scanning is quicker than parallel, it is not as faster as SCSI scanning.
|[ Team LiB ]|