Vol. XIV, No. 1

Spring, 2001

The last issue of the CSA Newsletter included an article about scanning slides to produce digital images ("Slide Digitizing -- A Lesson Learned," Vol. XIV, no. 1, Spring, 2001). The article described experiments carried out to compare the results of scanning with moderately priced slide scanners, sending slides to a local photo-finishing store charging about $1 per slide scanned, and sending slides to a more expensive lab charging about $4.00 per slide. The results indicated -- at least to CSA personnel -- that the difference in cost for the more expensive lab was justified.

The response to that article was quick, and readers suggested that more information and a more thorough report were needed. Therefore, CSA personnel undertook another, more carefully controlled experiment to provide fuller results for comparison, this time with three moderately priced scanners. A slide that had already been scanned at Luna Imaging and the local photo-finishing lab was chosen for scanning. It was scanned in each of the three scanners available on the Bryn Mawr College campus, a Minolta Dimage Scan Speed F2800 (with maximum resolution of 2820 d.p.i. and version 1.0 of the Minolta software), a Nikon LS 3510 AF (with maximum resolution of 3175 d.p.i. and version 4.5 of the Nikon software), and a Polaroid Sprint Scan 35 Plus (with maximum resolution of 2700 d.p.i. and version 4.5 of the Polaroid software). Unfortunately, documentation was difficult to locate; so it is impossible to be certain that each machine was operated at the best possible selection of settings.

The slide was scanned twice in each scanner, once with the default setting and once with the scanner adjusted so that the density range in the image matched the scanner's density range. Both scans were made at maximum resolution and maximum color definition. No adjustments of individual colors were made. The scans were saved as TIFF files and then transferred to the CSA computer. Each of the slides was opened with PhotoShop®, and the following operations were performed. Images were reduced, if necessary, to 24-bit color; they could not otherwise be saved in JPEG format. Two 300 x 300 pixel areas, one in the upper left (frieze) and another nearer the center (column and stylobate), were copied and saved as individual files. These files display the output of the scanners at maximum resolution, and each is small enough to be used on the Web. The files were saved in JPEG format.

Then each of the original files was reduced to one sixth (linear) of the original size to create images small enough to be displayed on the Web conveniently. (Note: In order to make the reduction simpler and to prevent problems that might arise, each file was first cropped, if necessary, to make the starting size - both length and width - a number of pixels evenly divisible by six.)

In order to include the scans from the photo finisher and Luna Imaging, the scans already obtained from them had to be adjusted similarly. Comparable details were produced from each, although the lower resolution of the photo finisher's scan made the images substantially different. Each was also reduced for display on the Web. The lower resolution of the photo finisher's scan permitted a 1/3 reduction (linear) in size, making the image nearly the same size as the others. Of course, the image has been reduced less; so it should appear to be sharper; nonetheless, this seemed a fair approach.

When the images were examined as a group for the first time, it seemed that the commercially prepared ones needed to be adjusted to be compared fairly. Each original scan was modified with the PhotoShop autolevel
command, and both details and complete, reduced-resolution versions were produced from the modified scans as well as the original scans. (The autolevel
command should have nearly the same effect as matching scanning density to slide density.)

The following images were available for comparison at the end of the process, not including the original, full resolution TIFF images and the photo finisher's PhotoCD image. Displaying them for useful comparison is difficult, and, in fact, no good mechanism for that has been found. The best choice seems to be to provide each image in its own window so that users may group and arrange them according to their own needs and desires. Therefore, each image will open in its own window.

Scanned Images for Comparison
Source	Full-view at lowered resolution		Detail at full resolution
			Frieze		Column and stylobate
Minolta	default	density- matched	default	density- matched	default	density- matched
Nikon	default	density- matched	default	density- matched	default	density- matched
Polaroid	default	density- matched	default	density- matched	default	density- matched
Luna Imaging	default	after autolevel command	default	after autolevel command	default	after autolevel command
Photo Finisher	default*	after autolevel command*	default	after autolevel command	default	after autolevel command
* images reduced to 1/3 (linear) rather than 1/6 original size

Readers will want to judge for themselves, but here are some observations from CSA personnel. The full images were different from one another in rather minor ways. Color casts were more apparent than differences in sharpness, and it seemed all would suffice for Web display. It might take some time and experience to become proficient with the scanners and software, but it seemed that eventually one could produce good results for this kind of reduced-size image from any of the scanners.

The two sets of detailed images were examined separately, first those showing the frieze course of the Propylaea and then those showing the column and stylobate. In each case, the better image for each scanner was selected, yielding five images to be compared. Then each of the five was compared to the others.

For the first group, the detail of the frieze, there was a difference of opinion, with the Polaroid scan and the Luna Imaging scan being preferred by Susan Jones and the Luna Imaging scan by Harrison Eiteljorg, II. There were differences as to color cast, but those were not the crucial reasons for preferring some of the images. The others seemed less sharp. Every effort was made to focus on the usable detail visible in each image, but the process of comparing the images was a necessarily subjective one.

Examining the second set of details yielded slightly different results. The Minolta image had washed out in areas; so it seemed less desirable than the others. At the other end of the spectrum, both Ms. Jones and Mr. Eiteljorg preferred the Luna Imaging result.

It would be preferable if there were some more scientific way to judge these images. It would also be helpful if there were some precise measure of the importance/value of a better scan when making archival digital images. The only conclusion here seems to be the one reached last fall. The Luna Imaging scan is better - or at least more uniformly so - and, for archival purposes, it seems to be worth the cost difference. Were these images to be used only for Web display at reduced resolution, that conclusion would certainly be different.

For other Newsletter articles concerning the The Propylaea Project, electronic publishing, or the use of electronic media in the humanities, consult the Subject index.

Next Article: Retrieving Dimensions from Drawings and CAD Models

Table of Contents for the Spring, 2001 issue of the CSA Newsletter (Vol. XIV, no. 1)

Table of Contents for all CSA Newsletter issues on the Web