The following is a guest post by Carl Fleischhauer, a Digital Initiatives Project Manager in NDIIPP.
This is the final blog on the topic of informational and artifactual values in the digitization of books (and other documents) and photographic negatives and transparencies. Here are links to the book-related blogs: Part 1 and Part 2.
Part 1 of the photo-negative blog ran yesterday, and began with my summary of the recent study by Don Williams, Michael Stelmach and Steve Puglia. The authors highlight the distinction between artifactual and informational in the case of digitized photographic negatives and transparencies. More important in practical terms is their goal of establishing a method to ensure success that “the full information content of the original scene” is captured when scanning.
How do we know what constitutes the “full information” of a given image? The authors used methods developed in the field of image science, many of which are specified by the International Organization for Standardization. One key measure is called Spatial Frequency Response. One method for measuring SFR employs a target (you shoot an image of it) that includes what is called a knife edge, a sharp boundary line between dark and light. This concept was applied to actual features in a real picture instead of a target, for example, the point at which architectural elements in a photograph of a building mark the boundary between sun and shadow. (There are lots of other wrinkles in this measurement game, fully described in the report).
The study’s results showed variation between classes of negatives. For example, the image information in an 8×10-inch glass-plate negative from 1906 was found to have a “limiting resolution” (the maximum information) of about 800 pixels per inch. In contrast, the limiting resolution on a 4×5-inch safety film negative from 1966 was about 2000 ppi. (In the intervening sixty years, the quality of lenses and film stock improved considerably.) In order to successfully scan these items, however, you must use equipment with even higher levels of resolving power. To be sure you captured the 800 ppi of the 1906 negative, your scanner must be capable of delivering a proven 1200 ppi.
One implication of this study is that many negative-scanning projects to date have been working at nominally higher resolutions than are needed to capture image information. It is also the case, however, that many such projects failed to use careful, standardized measurements to determine the actual levels of resolution delivered by the systems in use. “Your scanner will lie to you,” we sometimes say, “the 300 ppi it claims to give you is an exaggeration.” The pixels may be in the image but the underlying SFR metrics show that some are, um, empty–you aren’t getting 300 “real” pixels.
Meanwhile, the study highlighted two categories of historical color transparencies in which the microstructure was of interest to users. These are Autochrome and Dufaycolor photographs, created using special color processes in play early in the twentieth century. The Library of Congress holds a number of Autochromes, including this 1934 scene in or near the ancient city of Petra in Jordan. And a handful of Dufaycolor images, including a pre-World War II portrait of a woman in Ramallah in then-Palestine.
The image layer in an Autochrome plate consists of dyed potato starch granules that give the image a pointillist character, i.e., a visual texture reminiscent of the paintings of Georg Seurat, Paul Signac, and others. For researchers and other end users, this is an important aspect of this art form. For this reason, some Autochrome scanning projects seek to capture the microstructure–the artifactual aspect–as well as the image information. The research team determined that the starch granules occurred at about 1,700 per inch and they determined that a safe scanning resolution would be about 2500 ppi.
Thus we see different recommended scanning resolutions for two works from about the same time period: the glass plate negative and the Autochrome. In one case 800 ppi, in the other 2500. It makes a big difference when you seek to capture artifactual values.
At the July meeting of the Federal Agencies Still Image Working Group, we discussed the report and there was a helpful back and forth with a representative from the National Gallery of Art. Their scanning goal is artifactual: “The curators want to be able to see the cracks in the paint. When scanning paintings, the representative explained, the NGA captures at the resolution limit of the camera. Their current model will successfully cover items up about 16×20-inches in size. If the painting is larger than 16×20 they produce a set of “tile images” (segments of the painting) that are subsequently stitched together in software. The NGA has found that when publishers lay out their fine art books, they always request some unexpected small detail within a large work of art. Meeting this objective drives the need for very high resolution imaging.