I am using a Hewlett Packard ScanJett IIc, about 800 dpi with some interpolation, color / bw. This is an "older" model, so I bought a used one for 500 sFr. (swiss franks).
- OCR even for small printed text (as in journals) is satisfying with OmniPage.
- Microscopic slides *can* be scanned, resulting in about a 30 fold magnification.
Personally I am very satisfied with this device. Good Luck, Wolf.
wschweitzer@access.ch (Wolf Schweitzer)

K.-R. Peters wrote:
Optimal scanning of negatives requires that the negatives are exposed 1-2 stops more than for ordinary photographic printing (high lights should be "covered" and not be empty, dark areas should be very dark. Good scanners can handle more than 3.0 optical density which is "very dense" for a photographic negative). Also the scanning procedures must be adapted in order to take advantage of the full contrast transfer function of the negatives as well as the scanner. I am in the process of putting a report together at my WWW site on our experience with 8-bit versus 12-bit negative scanning.
Best regards Klaus
Klaus-Ruediger Peters, Ph.D.
e-mail: Peters@BSAC.UCHC.EDU
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
we have partially overcome this problem by overlaying negatives with photographic contrast filters or neutral density filters. This may not be appropriate for very high res. work but is no problem for video display and printing to a laser printer
Greg Erdos
E-mail: gwe@biotech.ufl.edu

K.-R. Peters wrote:
Optimal scanning of negatives requires that the negatives are exposed 1-2 stops more than for ordinary photographic printing (high lights should be "covered" and not be empty, dark areas should be very dark. Good scanners
Greg Erdos responded:
we have partially overcome this problem by overlaying negatives with photographic contrast filters or neutral density filters. This may not be appropriate for very high res. work but is no problem for video display and printing to a laser printer
Greg Erdos
E-mail: gwe@biotech.ufl.edu
Thanks Greg, you make an important point concerning acquisition and display:
The application of contrast or neutral filters in photography shifts the brightness (level) or contrast (slope) of the image data to a range at which the eye is more perceptive (details in high lights may be preserved, but reduced in the shadows or visa versa, or both may be compressed). These procedures do not produce more "image detail" than the negative contains but modifies the display of the details. If during image acquisition (negative exposure) the highlights in the negatives were underexposed (or the shadows overexposed) then the possibly lost detail information can not be recovered, neither analog nor digitally.
It is therefore important (for both analog as well as digital imaging of the negative) to the set the brightness level of the negative so that it captures as much information as possible. Generally, the contrast transfer function (slope) is not changed, i.e., through variation of developing conditions or change of film/developing combination. But why don't we normally care much about optimal exposures (or acquisition)?
Photography (analog) had long accepted that there is more information in a negative (raw data) than the eye can "see" ("image" being that part of the displayed image data which the eye can perceive and recognize). It developed many useful tools for enhancing the contrast of certain image components of the negative so that they become visible, but consider our following findings. Measurments of the intensity ranges of contrast patterns in digitized EM negatives and the contrast ranges required for visual display of these contrast patterns (as images) indicates that an optimally exposed TEM negative has an information depth of approximately 12 bit, or a 12-bit contrast resolution whereas the "image" has merely an approximate 6-bit contrast resolution. We find in digital image data from nearly all microscopies that the detail contrasts (structures of lowest contrast levels) occupy only 1-10 % of the overall contrast range of the image data whereas we need about 25% of the contrast range of an image for visual recognition (not perception) of a contrast pattern (in an amorphous environment). This means, most of the low-contrast data components (which constitute the high-resolution details) are not visible in a conventional image of the data. Unfortunately, we have adapted to this fact and tend to acquire images at an insufficiently low contrast resolution level equal to that of our visual system because we are not missing anything in the "empty" high-lights or empty shadows although our microscopes can acquire images at much higher contrast resolution. On the other hand, these facts underline why digital acquisition (from the imaging sensor or from the negative) should be done at a 12-bit level instead of an conventional 8-bit level. Many manufactures of acquisition and processing equipment start to provide a 12-bit "output" level, 12-bit scanners are now offered at catalog firms, and Photoshop is now available for PCs and Macs with 12-bit TIFF capabilities (version 3.04).
The basic idea of scanning negatives is to "preserve" all information of the negative with a linear transfer function, irrespectively if the eye on the negative or print can see it or not, and then use digital imaging for the display of all the available image information. Thus, when exposing a negative for digital scanning, we should capture the contrasts of interest with the largest possible contrast range. Since good scanners can penetrate the darkest black of a negative (max. OD = 3-3.5) capturing the details in the shadows, we should expose more than conventionally done in order of acquiring also as much detail as possible in the high-lights.
Additionally, one should consider that laser printing (HP LaserJet 4 providing 5-bit gray levels and 100 lines/inch) combined with the LazarPrint expansion can print 8-bit gray levels at 300 lines/inch, thus provides a good output for detail rich images at the 1Kx1K level (I tested recently several printers and reported the results in my home page).
Klaus-Ruediger Peters, Ph.D.
e-mail: Peters@BSAC.UCHC.EDU

In response to Klaus Peter's eloquent plea for 12-bit versus 8-bit scanners:
While I would not want to be thought of against precision, it must be admitted that there are some limitations to using 12-bit data, namely greater storage (and display?) requirements. So one should consider this step carefully.
Assuming that it is done correctly, 12-bit data is digitized to 1 part in 4,000 while 8-bit data is digitized to only 1 part in 256. Therefore, to make proper use of such precision one must have data of similar precision. Although many considerations can reduce the precision of recorded data: measurement noise, low contrast, it can never be better than the statistical limitations placed by quantum mechanics on the number of events counted (grains of silver (assuming that they are all of the same size which whey aren't)/pixel or photons/pixel or electrons/pixel). Clearly, this all depends on the size of a pixel: larger pixels imply the chance to count more quanta (but they also mean lower scanner resolution) so you may need less intensity resolution if you have greater real (non interpolated) spatial resolution.
Now it is clear that, for instance, in STM, where a 0.05 nm height resolution can be detected and the piezo may have a height sensing range of many microns, there is no difficulty in obtaining 12-bit data (or even 16-bit data) although you may have trouble displaying all of this "depth" to a human observer at one time. Likewise, CCD sensors having noise levels of +/- 3-4 RMS electrons/pixel and full-well signal levels of 300,000 electrons/pixel easily satisfy the 4,000:1 (more like 100,000:1 possible) requirement as long as you use a long enough exposure to actually record at least 20,000 electrons/pixel. However, in confocal microscopy, 256 photons/pixel is often quite a high signal and digitizing from a color slide of a fluorescent image recorded on high speed film (depending on original magnification and pixel size. Remember, 1200 DPI implies 20 micron pixels.) one may find it difficult to find even 10 distinguishable levels.
So far we have only spoken of linear digitization (as is appropriate to scanners which usually make every effort to be linear). However, for signals degraded mainly by statistical noise and recorded by the direct digitization of electron signals, the separation between "meaningful" grey levels (those separated in intensity from neighboring levels by an amount at least equal to their standard deviation?) the difference between the first two such levels (1 event/pixel and 4 events/pixel) is 3 event/pixel whereas that between the 15th level and the 16th is 31 events/pixel. In other words, if we are only interested in recording the INFORMATION in a signal limited only by quantum noise (SEM? Confocal?), we could first take its square root and then digitize this. This could be done using only one half of the number of bits that would have been necessary to preserve the information in a linear signal and each SQRTBit would be a "real" grey level.
All this said, one must admit that getting real 8-bit data requires more than the use of an 8-bit DAC (0.4% illumination stability, low-noise detector, freedom from digital electronic interference, proper bandwidth for sampling rate in both the electron and optical parts of the information path, etc) and many scanners may not provide the performance they claim. This might explain much of the difference claimed to exist between 8-bit and 12-bit results.
(To any owners who have read this far: have you any MEASUREMENTS on scanner performance on known images?)
SEM might provide a good case in point: The SE signal variation associated with small features is often less than 1% of the total signal. As "seeing" a small feature with only 1% contrast requires collecting at least 1/0.01 x0.01 = 10,000 electrons/pixel, it might seem worthwhile to use 12-bits to preserve these small variations. However, unless there are "holes" in your specimen from which virtually no signal is obtained (i.e. if your signal does not have "important" high contrast), little would be lost if one digitized only the 256 levels nearest in intensity to those of the small feature and a positive improvement would be gained (all the information that could possibly be coded in up to 65,000 detected electrons vs. 10,000) by digitizing the square-root of the signal into only 8 bits. In either case you would have to "process" the signal before you displayed it on the screen.
Just some thoughts. Sorry that they were so long but I felt a trend to Digital-One-up-man-ship..
. Jim Pawley
JBPAWLEY@FACSTAFF.WISC.EDU

To all,
I got a number of good responses on my question regarding flatbed scanners. Everybody seemed pleased with the particular one they owned--there were no clear favorites. A couple of tips: 1) Make sure your computer has the right output. All of the scanners I have looked at have a SCSI 2 output while my 2.5-year-old Mac only has a SCSI. I have been told that I have to upgrade my computer in order to even hook it up to a SCSI 2 scanner. I'm still investigating this. 2) Make sure you get the software you need to manipulate the scanned image or text. A lot of scanners come with software bundles. 3) Look into warranties. Some devices have none--others are up to 2 years (longer?). 4) Photoshop comes in two versions--LE and 3.1. The latter is more powerful (and expensive) than the former. 5) Transparency adapters and document feeders are usually extra. Both can be useful.--if you need them. 6) Read the first response (from ). It contains a lot of good basic advice. 7) Good luck.
Bob
________________
From: sco.umc2@Mail.health.ufl.edu
One user's opinions:
1. Consider the media you will be scanning. Some scanners do better at transparencies than others (nikon has a slide scanner for 3K$), some are designed for large format (Umax has a 17 x 12 inch for 7k$), some do black and white on the cheap (HP has one B&W for $400).
2. Consider the destination of the image. There is no use buying a 30 bit high definition scanner if all you will do is newsprint. A 300dpi 256 gray scanner goes a long way with laser printers and costs $400.
3. Consider the resolution you need. If you are going to scan stamps that are 1 inch, and display them at poster size, you need a very high res scanner (1200 dpi or more) if you are going to scan 8 x 10 positives and display them in a small window, then 600 is plenty, 300 adequate.
4. Consider the speed. Some scanners take 30 seconds to scan one page. This can be a real pain if you have any number of images to scan. Some are available with a document feeder, which is a nice thing to have, but not if all your scans will be of pictures (which must be hand fed anyway) "Single pass" scanning means that all 3 colors (red green blue) are captured simultaneously--this is faster (HP scanjet 4 offers this)
5. Real Res. Most scanners interpolate the dpi at high res. This is ok, but non-interpolated scans are better. Be willing to pay more for a hardware 600dpi than a software 600dpi.
__________________
Sender: bob@befvax.uchicago.edu
What resolution do you need. Many of the inexpensive scanners have a very small numerical aperature which leads to line broadening. So a step size of (for example) 5 microns will not resolve a 5 micron line in single (or two) pixels but rather 10 or more. The requirement for a large numerical aperature coupled with dimensional stability is what makes high resolution cost so much.
Bob J.
______________
Response from John Bozzola:
We use a Mac IIci attached to a LaCie 600 dpi color scanner (it's equivalent to an Epson 300) to occasionally digitize prints. It cost $1,800 - a similar unit may now be purchased for around $600. I have used it for over 3 years now with no problems whatsoever. 97% of the scanning is done in the bw mode set to no more than 300 dpi. The program most often used to drive the scanner is Adobe PhotoShop. As an OCR program, I use WordScan Plus. It's OK, a bit slow, but it is also 3 years old and a newer version may be better. I rarely use this program. However, when the documents are of high quality (typed characters no smaller than 9 point) then it is a great time saver. There are better OCR programs out there than this one.
_________________
Our Graphics Artist uses a Macintosh computer with a Hewlett Packard scanner which he likes very much. The one we have is color and has a 400 optical DPI, although the new Hewlett Packard has a 600 optical DPI. It comes with Deskscan software and a limited version of PhotoShop. If you want to use color, though, you need to have the full version of PhotoShop.
Hope this is helpful.
Kathy Stangenberg
Ted Pella, Inc.
_______________
I have a AGFA StudioScan IIsi scanner with Transparency module. I think that it is a very good scanner. I using it every day. Some technical data:
- Max. res. 400(H) x 800(V) optical 2400(H) x 2400(H) ppi through interpolation

Sample depth: 10 bits for grey 30 bits for colour

Scan mode: one pass

Scanning speed: grey 4.5 ms/line colour 10 ms/line

Scanning area: 316x355 mm (8.5"x14")

                           MAC              PC

                        SM      QM      SM      QM

Preview colour          16      16      15      15

A4 colour 200 ppi       54      54      32      39

A4 colour 400 ppi       96      104     86      126

15x10 cm colour

    400 ppi             44      54      30      38

A4 grey 200 ppi         17      18      12      12

A4 grey 400 ppi         39      42      39      65

A4 line art 400 ppi     20      34      20      34
           

I was leafing through Macworld the other day and realized that they had
reviewed several scanners in the last 6 months.  The recommendes ones were
(4 or more stars out of 5):

Flatbed scanners:

HP Scanjet 3c
        30 bit dynamic range, 600 dpi optical resolution, list $1179,
street ~$700
        208-323-2551

PixelCraft Pro Imager 8000
        30 bit, 1200 dpi optical, list $13k
        510-562-2480


35 mm negative scanner:

Polaroid SprintScan 35
        30 bit, interpolated 2700 dpi, list $2495, street ~$1600