30,000 pixels on a side, and uses virtual memory on disk to handle images larger
than will fit into memory). A digital camera connected to a macro lens can certainly
achieve a spatial resolution of 2000 dpi, but over a much smaller distance of an inch
or two. To image such a large area would require a corresponding reduction in spatial
resolution. So for situations in which images of sections through food products can
be obtained by cutting them and placing them on a flatbed scanner, doing so provides
an extremely high quality result at a very low cost and with great convenience.
Figure 2.2 shows a representative application.
With a scanner, there is no concern about focusing (provided the sample is
relatively flat) or achieving uniform lighting. The hardware takes care of that
(although avoiding the very edge of the platform, where light intensity may fall off
slightly, is a good idea). There may be some clean-up problems, but these are at
most a minor nuisance. The time required for scanning is typically tens of seconds,
but that is usually fast enough. Scanners come with a variety of standard interfaces,
ranging from older SCSI (small computer systems interface) to the newer, faster,
and more trouble free USB (universal serial bus) or IEEE 1394 (widely known as
firewire). Many scanners are bundled with software such as Adobe Photoshop that
provides a good platform for image acquisition, storage, printing, and some of the
processing techniques discussed below and in subsequent chapters.
Scanners also typically have a high dynamic range. This is usually specified as
bits of information in each channel. The scanners have sensors that read the red,
green, and blue light intensity separately. This is typically done by using one or
several linear sensors with colored filters that are mechanically scanned across the
area. The maximum dynamic range is determined in part by the well size of the
detectors, that is, the number of electrons that can be accommodated, and the noise
level of the electronics used for readout. The readout is much slower than a typical
digital camera (and considerably slower than a video camera). This results in less
noise and a higher dynamic range.
In any case, the individual red, green, and blue light intensities from points on
the sample are digitized — converted to a numerical value — for transmission to
and storage in the computer. An 8-bit image, for example, uses one computer byte
to store each of the RGB channel values, and since 2
8
is 256, it can represent 256
discrete brightness values in each channel. That may seem like a lot, and for some
purposes it is. Display of the image on the computer screen for human viewing, and
printing the image out as hardcopy, does not require more than 8 bits and can be
adequately accomplished in most cases with less. But for many of the image pro-
cessing operations to be described, and to detect small variations in brightness that
may represent local structural details in an image that has a large overall contrast
range between bright and dark values, it is best to have more tonal resolution than
the 256 values provided by 8 bits. Indeed, many professional photographers suggest
that all work on digital photographs be performed in a 16 bit space, only reducing
the image to 8 bits for the final printout (because printers cannot handle and do not
need the increased amount of information).
Actually, few detectors and scenes have 16 bits of information (2
16
= 65536).
Photographic prints do not typically have as much as 8 bits of data, although the
negatives can be much better as discussed below. In astronomy, where stars are very
2241_C02.fm Page 53 Thursday, April 28, 2005 10:23 AM
Copyright © 2005 CRC Press LLC