Is Bigger Better? Web Delivery of High-Resolution Images from the Museum of Fine Arts, Boston
Perhaps one of the most frustrating things about working on the content side of a digital library is knowing how many resources in the library are not yet published and available to the world. Like a museum curator with limited gallery space and a storeroom full of artworks, a digital librarian is always thinking of the day, some time in the future, when the unknown parts of the library will have their turn in the limelight. This is why we are particularly happy to be able to present the work on high-resolution images of Roman coins which the Perseus Project (http://www.perseus.tufts.edu) has carried out in collaboration with the Museum of Fine Arts, Boston (http://www.mfa.org), with its curatorial staff in Art of the Ancient World, John Herrmann, Mary Comstock, and Pam Russell, and the museumís Director of Information Resources, Nancy Allen. This collaborative effort with the MFA forms part of the continuing research and development at the Perseus Project, an effort now in its thirteenth year, at Tufts University. Our work represents another major step toward the electronic delivery of scholarly-level museum resources in a permanent, standardized, flexible way.
The limitations of electronic resources, especially for scholarly use, are well known to all of us; I outlined some of them in another Museums and the Web conference paper (Daniels, 1997). At that Los Angeles session, Prof. Charles Rhyne of Reed College made explicit the particular problems of teaching art history using inadequate visual materials, and stressed, as he has in many of his publications, the necessity for scrutinizing technical details, such as brush marks, or small parts of an object, like the dozens of tiny figures on a 12-foot-long Japanese screen (Rhyne, 1997). We agreed that scholars, students, and teachers would not truly be able to take advantage of the electronic environment for their work until more, higher-resolution images were made available. At Perseus, we had always concentrated on the quantity of images we included, up to 109 detailed views of a single Greek vase . However, in the back of my mind was the annoying knowledge that the Perseus Project did, in fact, have an archive of scanned and digital camera photographs at a much higher resolution than the minimal, 640x480 pictures we freely published on our web site. What was impeding the distribution of images of a significantly higher quality? Could these high-resolution pictures be the analytical tools our users still lacked?
Later that year, as we embarked on the Perseus-MFA collaboration that continues today, the question of delivering higher-resolution pictures was foremost. Work over the past two years has culminated in the electronic publication of a major catalogue of Roman art, including 780 Roman coins and medallions, with a series of Perseus enhancements including a tool enabling viewers to zoom in on 18-megabyte photographs, examining the coins at up to 25 times their actual size. This paper will begin by discussing the principles which guided us, then continue with a short description of our working methods. Finally, it will venture to suggest some of the earliest results of this work. The question of the title, "Is Bigger Better?" is really a question for you, the audience of museum professionals, art educators, and everyone interested in electronic access to cultural resources. How can these big images be integrated into an electronic museum experience, and what potential do they hold for the transformation of museum collections into complete educational resources?
As Skidmore & Dowie (1999) point out, educators implementing technology do best to begin by articulating the educational objectives which need to be served. At the outset of our work, members of the team, including curators, museum educators, technical staff, and university faculty, worked together to clarify the educational goals of this project, and, further, to express a set of principles which would apply not only to the instructional uses of the data, but also to its longevity and utility over time. It was particularly important that the collaborators started with a discussion of the projectís objectives; otherwise, differences between the museum and the university perspectives might have led to difficulties, had we failed to articulate our individual interests and work together to integrate them into a single approach at the outset.
The first, and most important, principle guiding our work with the MFA coins was the understanding that these objects are transformative. A coin portrait of a Roman emperor like Trajan, a depiction of the Mausoleum of Hadrian, a personification of Gaul being speared by a Roman warrior, issued during Caesarís conquest of Gaul ó all of these representations have the power to change the way people understand the ancient world. Yet I am not making a special argument for Roman coins, even though coins are among the hardest to view in a gallery setting, the most challenging for curators and educators to label and explain, and the easiest to overlook when they are surrounded by a building full of larger, more emphatic art works. At the outset of our collaboration, we agreed that the experience of seeing a Roman coin could be enlightening, and we therefore included coins in our plans to digitize the Roman collection. We also included glass, pottery, mosaics, jewelry, portable altars, mirrors, statues, tomb reliefs, clothes pins, pieces of buildings, and other objects. Our approach, in short, was that if an object is worth collecting, it is worth digitizing. This principle was tested by the usual limitations of time and money, of course, and we used a few broad themes, such as daily life and Roman gods, to help direct our selection of objects. However, we avoided a "greatest hits", "sculpture only", or other such limiting approach to the collection, as much as possible, opting instead to represent as wide and as deep a sample as we could. The work presented here on coins is being extended to all 1,055 of the art works we have documented thus far.
A second principle was that of integration, not only of the different sources of information within the museum, but also of the newly digitized material with existing resources at the MFA and in the Perseus digital library. Written documentation for the Roman coins came from extensive curatorial records which had not yet been added to the museumís collections management database, and the museum insisted that this digitization project, and all others, should produce data useful for their main database. On the Perseus side, we were hoping to find added value for the new objects by placing them in context with the other holdings in the Perseus digital library, including photographs and descriptions of Roman sites, an atlas of the ancient world, and the texts of Cicero, Caesar, Josephus, and other authors. One other consideration was the integration of several existing publications of the art works, where possible, in order to avoid reduplication of efforts, and to bring new users to the scholarship already in publication. The only exception to this integrative approach was the creation of new photographs, which were deemed necessary in order to provide consistent coverage of the collection. Many objects had never been photographed in color, or never had detail photographs made; the coins themselves were all photographed with a centimeter scale, in order to convey their relative size.
A third underlying principle was one which hardly needs elaboration here: control of ownership. Everyone involved agreed that protecting these new resources, by asserting copyright, watermarking data, and attaching ownership and credit information to the photographs, was important. From the start, our goal was to publish the catalog of coins on the World Wide Web, and make it freely available to the general public. However, we chose to implement several safeguards, to ensure that the pictures would not be separated from their descriptive data. In the usual Perseus collaborative arrangement, the MFA retained ownership of the data we produced, in exchange for allowing it to be published freely on the Web. Thus, we had an obligation to protect the data, even as we sought for it a wide general audience.
Several more principles guiding us in this work related to the data collection process in a more technical way. First, we determined to adopt and conform to existing standards, for the data itself and for its structuring, as much as possible. Controlled vocabularies, standardized spellings, TEI-conformant SGML tagging, standard image file formats and processing, and compatible database structures at the MFA and at Perseus all played roles in making our data coherent, and in allowing this work to scale successfully. The curators invested their time and expertise in determining whether the quality of existing documentation was adequate, and, if not, creating new catalog information that would meet their standard. Time spent building tables of credit lines for objects, compiling alternate names for the same representation (e.g. Castor and Pollux, the Dioscuri, marking up bibliographic references, and employing standard vocabulary sources, like the Getty Art & Architecture Thesaurus, contributed to the robustness of the data, and ultimately saved some effort in the processes of data entry and production for the web.
It is worth noting that in some cases, standard classification systems have not yet developed to include the highly specialized data sets relevant to our work; as Jörgensen (1999) indicates in her thorough survey of image retrieval systems, it is still a struggle to understand how people search for images, let alone to build adequate generalized systems to aid those searches. An example is the Getty Thesaurus of Geographic Names, which does not yet include all of the geographic data from the Mediterranean which would be useful in classifying the assorted geographic attributes of the art objects, including their findspots, depictions of places, inscriptions naming places, or the sources of the objectsí raw materials. To address this problem, we have sought help from classification schemes of other specialized sources, including field-specific publications and other archives of ancient art, and in some cases developed structures ourselves, based on the Perseus-specific content.
A second, critically important technical principle was longevity of data. One of the paradoxes of digitizing two-millennium-old art works was the knowledge that they themselves were likely to be legible long after our databases and our digital cameras crumbled into dust, if we failed to make the data portable. While the Web is the immediate delivery mechanism for our work, all of the data is structured so that it can be republished in emerging formats for years to come. The Perseus Project has a long experience in developing this type of portable data; for almost a decade, the project developed back-end data in powerful relational databases and structured SGML, but delivered the material on CD-ROMs in a low-cost, widely available HyperCard front end. With the advent of the Web, we were able to take this same data and port it to a Web front end within a matter of weeks. We have also successfully built delivery software for the same data for a platform-independent version of our CD-ROMs (Crane 2000).
The last technical principle we kept fixed firmly in mind was a focus on automating the digitization process. So much of the work necessarily had to be done by hand: the positioning and lighting of each object for photography; the review of catalog information, and writing of new materials; the checking to ensure images were correctly attached to the corresponding objects. Yet in every possible instance, we looked for automated tools which might speed the process, from batch image-processing software to database error-checking scripts to automatic tape backup.
Our final two requirements as we commenced this work were seemingly disparate principles that have proven over time at Perseus to be actually quite complementary: we sought to promote scholarship and also to distribute our work to the widest possible audience. In pursuit of the former goal, we adopted the highest academic standards in compiling digitized resources; in pursuit of the latter, we recognized that the distribution medium for our work already far outstrips the distribution network for traditional scholarship, and labored to provide the interface, organization, and tools that would facilitate the materialís usefulness to a broader range of users.
The range of testimonials we constantly receive from site visitors, preliminary analyses of Web logs, and the sheer quantity of page requests the site receives (currently hovering around 200,000 page requests per 24-hour period), together indicate that a digital library has the demonstrable ability to simultaneously serve scholarship and reach an emerging broad audience. We will be able to evaluate site usage, to learn more about our audience for the high-resolution images; for instance, by comparing image use at Internet 1 and Internet 2 schools, we will be able to investigate how having the infrastructure for large downloads will affect use of those resources.
To summarize, the principles informing our approach to this work included: a broad inclusion of objects; integration of resources; control of ownership; adherence to standards; emphasis on data longevity and portability; automation of the digitization process; promotion of scholarship; and, finally, acknowledgement of a wide audience in the general public.
From the principles we developed a fairly streamlined work process that has taken advantage of many current standards in digital library work. In addition to the standard classification systems and data formats outlined above, we chose a range of software and hardware with a preference for robust, open-source solutions. Written documentation was produced in a FileMaker Pro database designed to be compatible with both the Museum of Fine Artsí collections management database, in FileMaker Pro, and with the Perseus art and archaeology databases, in 4th Dimension. After completion, this data was transferred into PostGres, an open-source SQL database program running on a Linux server, which allows us to run certain indexing functions, automatically generate cross-links with the rest of the digital library, and publish the data on the Web. Thanks to these automated functions, Perseus can immediately associate terms in the new coin documentation with existing terms in the digital library, and lead curious readers to a wide array of additional information; for instance, when a site like Ostia is depicted on a coin, the word "Ostia" in the coinís description automatically links to other Perseus resources on Ostia in the Lookup Tool, including an article on Ostia in the Princeton Encyclopedia of Classical Sites, links to other objects depicting or coming from Ostia, a set of pictures which include "Ostia" in their captions, an atlas page that will plot Ostia on a map, and bibliographic sources on Ostia.
On the imaging side, the only stumbling points in a very smooth process have been, first, the proprietary format of the digital photographs as the camera initially stored them, and second, the acquisition of enough hard disk space and RAM to manipulate, store and deliver these quantities of visual information. Digital photographs were made using a Kodak DCS 460 camera, chosen for its relatively high resolution images, its compatibility with our existing system of Nikon lenses, and its efficient work flow, with sufficient storage on removable cards. Once the images were captured and backed up, they were converted to standard TIFFs, a time-consuming process run under the Photoshop Acquire mode which would have been greatly assisted by automated conversion software; we were unable to develop this tool in-house, and Kodak did not produce it, but fortunately a third party vendor, DSL Consulting Inc., has written an AutoAcquire script that finally addresses this problem.
Once the images were converted to their delivery format, we employed an assortment of identification and marking measures. Although current watermarking programs are not impervious to attack, they are a perfect mechanism for linking museum identity with each image. As Peticolas & Anderson (1999) explain the current technical outlook, "although most [watermarking] schemes could survive basic manipulations they would not cope with combinations of them or with random geometric distortions." Still, visible and invisible watermarks are demonstrably useful tools that can be employed first, to link ownership information to image files, and second, to ensure that each image is difficult to separate from this identifying metadata.
In order to deliver large files across the net in a reasonably short time, we chose to tile the coin files into scrollable windows (http://www.perseus.tufts.edu/cgi-bin/zoomer?lookup=1997.03.0019); the largest tiles are c. 475K in size and compress down to about 36K as JPEGs, due to their limited color palettes. Like GridPix, a tool developed at Berkeley, and other tiling programs, the Perseus tiler allows users to interactively zoom in to parts of an image and scroll around it, while minimizing the waiting time for tiles to appear. We are constantly reminded that many visitors to Perseus dial up over slow modems, log in from geographically far-flung places, and compete for bandwidth with hundreds of others using the same service providers. To speed access, we generated all the different sizes and tiles in advance, not on the fly, and stored the thumbnail size, small (600x400), medium (1530x1018), and large (3060x2036) versions of the images in parallel directories on our server. A database keeps track of what resolutions exist for each image. For implementation, we first used standard archiving software to retrieve the high-resolution, 18 megabyte digital photographs from a tape archive, then employed open source image packages, including PNM and ImageMagick, to convert them into their assorted derivative images.
An important innovation of the Perseus tiling program is a random rotation feature, which delivers each tile slightly rotated a random amount either right or left, within a narrow range. While not affecting the look of the displayed image, the random rotation protects the tiles from re-use, by making it very difficult to reconstruct them into a single, high-resolution image file.
A late improvement to the scrollable image window has been a side-by-side image display tool, still in the final stages of development, which allows for simultaneous display of two images at the series of available resolutions. We strongly agree with commentators like Skidmore and Dowie (1999) that side-by-side image presentation is a core pedagogical approach in art history that has been sadly ignored by new technologies up until now. In a 1997 evaluation, one of Rhyneís students commented that a setup presenting two images provided "distinct advantages" for her study, making "manipulation and locationÖeasy" and facilitating "side-to-side comparisons impossible in the museum." Her summary opinion: "The scant hour of frantic sketching and note-taking in a crowded room has few advantages over these clear, beautiful, easily maneuvered and compared images which students can investigate individually." Our intention is that the Perseus image display tool will serve as a vessel for teachers to construct and save sets of images in Perseus, just as they now save sets of slides for the left and right carousels of their slide lectures. This tool, like other Perseus resources, has been built generically, so that it will be able to handle a variety of image inputs, from coin pictures to page images of texts to custom maps.
How will access to these big images affect research possibilities for a broad audience? We feel much the same curiosity expressed by Schwartz (1999), but enough confidence in what we have built to be able to say we have indeed "provided tools for accessing more (or more meaningful) information than was previously available." The museum staff are pleased to have created a new public platform for the scholarship on so many objects in their collection, which has until now found only a specialized audience. Usersí first responses to the image display tool and the high-resolution images have ranged from delight to glee. Proper evaluation, both by internal groups of teachers, museum educators, and advisors, and external evaluators, will take place as part of a grant from the Digital Library Initiative, Phase Two, from the National Science Foundation, the National Endowment for the Humanities, and other government agencies. With this support, the Perseus-MFA collaboration is also continuing, and the ongoing focus will be to build generic digital library tools for the humanities which will be made accessible to the wide audience. This set of tools is applicable not only to works of ancient art, but to other humanities sources, such as a First Folio of Shakespeare, a Renaissance Italian dictionary, the physical remains of ancient Rome, and a series of Giza mastabas replete with hieroglyphic inscriptions. Visitors to the Perseus web site will have the ability to examine high-resolution images of any of these resources, using the same image tools developed for the Roman art of the Museum of Fine Arts.
We may not be able to judge right away whether, as Schwartz asks, "the site provide[s] a window of introduction that will make the use of the museumÖmore meaningful," even if it might seem clear that giving visitors an improved capacity for scrutinizing art in a way that is difficult to do through the vitrines, under the spotlights, or among the crowds in the galleries, would necessarily bring more nuance and meaning to their understanding of the works they see. My favorite story on this subject is the true one of the Perseus web site user who printed out catalog entries from the exhaustive Caskey-Beazley publication of Greek vases in the MFA, came to Boston, and walked around the galleries with printouts in hand, gaining appreciably more information from looking at the objects in tandem with the thousand-word curatorsí essays than he ever could have from the 300-word labels.
Yet even with insufficient evidence to judge whether the physical museumís use becomes more meaningful, we can be certain that the museum has been transformed into a place now accessible, for the first time, to a new constituency. Virtual visitors, who might never travel to Boston, can now see the collection from the Ivory Coast, from Brazil, from Italy, from Singapore, from other parts of our own country, and from hundreds of remote places. These visitors embody the same attitude Dierking & Falk (1998) have observed in actual museum visitors: they "profess high to moderate interest in the subjects presented" they have sought the subject on the Web ó but at the same time profess "low to moderate knowledge" they have come looking for information. The virtual museumgoers already demonstrate that providing on-line documentation of the museumís resources, particularly high-resolution images, transforms the museum and its collection from a local storehouse open 60 hours a week to an internationally available educational resource open 24 hours a day, allowing more people to have an in-depth experience with the art works than ever before, and democratizing the opportunity for a thoroughgoing educational involvement with the collection.
In closing, I would like to thank Charles Rhyne for his challenging questions three years ago at this conference. I titled my paper with a question, not because I can provide a simple answer, but because I hope the collaborative work of the Perseus Project and the Boston Museum of Fine Arts will prompt your own questions, comments, and ongoing discussion about how high-resolution images can change our way of learning with art, by enhancing the way we are able to see it.
Crane, G., Jacob, R., Taylor, H., Scaife, R., and Allen, N. (1998). A Digital Library for the Humanities [online]. Funded proposal submitted to the NEH-NSF Digital Libraries Initiative, Phase II. Available: http://www.perseus.tufts.edu/Props/DLI2/dli2.html. (February 15, 2000).
Crane, G., ed. (2000). Perseus 2.0: Interactive Sources and Studies on Ancient Greece. Platform-Independent Version. New Haven: Yale University Press [forthcoming].
Daniels, M. (1997). A Wish-List of Web Resources for Humanities Scholarship [online]. Paper delivered at Museums and the Web 97. Available: http://www.perseus.tufts.edu/~maria/MW97.html. (February 15, 2000).
Dierking, L. & Falk, J. (1998). Understanding Free-Choice Learning: A Review of the Research and its Application to Museum Web Sites. In D. Bearman & J. Trant (Eds.) Museums and the Web 97-99: Special Edition Proceedings. CD ROM. Archives & Museum Informatics, 1999.
Jörgensen, C. (1999). Access to Pictorial Material: A Review of Current Research and Future Prospects. Computers and the Humanities 33, No. 4, 293-318.
Rhyne, C. (1997). Images as Evidence in Art History and Related Disciplines. In D. Bearman & J. Trant (Eds.) Museums and the Web 97: Selected Papers (pp. 347-361). Pittsburgh: Archives & Museum Informatics, 1997.
Rhyne, C. (1997). Student Evaluation of the Usefulness of Computer Images in Art History and Related Disciplines. Visual Resources: An International Journal of Documentation, XIII, No. 1, 67-81.
Petitcolas, F. A. P. & Anderson, R. J. (1999). Evaluation of Copyright Marking Systems. Proceedings of IEEE Multimedia Systems í99, vol. 1, 574-579.
Schwartz, D. (1999). Museums and Libraries in the Age of the Internet: Lessons Learned from a Collaborative Website. In D. Bearman & J. Trant (Eds.) Museums and the Web 97-99: Special Edition Proceedings. CD ROM. Archives & Museum Informatics, 1999.
Skidmore, C. & Dowie, S. (1999). Camera Lucida: AMICO in an Art History Classroom. In D. Bearman & J. Trant (Eds.) Museums and the Web 97-99: Special Edition Proceedings. CD ROM. Archives & Museum Informatics, 1999.