When I began this research I had formulated the concept that image databases should have a philosophical outline. Such considerations encompass some interesting questions. My select two follow:
Will this image collection support textual information to illustrate the political content and cultural colloquialisms that the future will not be able to interpret without this information?
Or will the database be able to allow this kind of search as well as a less informative one in which the user will be able to formulate his or her own ideas about what would now be history?
I decided that before approaching these questions I should investigate the process and progress of image retrieval applications. Now, weeks later, I see my questions overflowed with idealistic vision contained within the ignorance of misunderstanding the parallel journey between technological advancements and the basic requirements of information retrieval (particularly image retrieval).
This essay is directed towards those like myself, who have already completed a wishlist for image databases without understanding the past restraints. It is meant to briefly examine and compare the access needs of visual collections confined within technology's capabilities, complementing the human effort and insight that has initiated the tremendous progress in database systems.
In Helene Roberts article, "Visual Resources: Proposals for an Ideal Network (pp.32-33)," she re-accounts a method of using visual aids to support narratives of visual argument used by art historian Aby Warburg in the 1930's. His assistant, Fritz Saxl invented fabric-covered wooden frames in order to display an array of images in different arrangements to illustrate the research of Warburg. Eventually the screens numbered to forty, portraying over a thousand moveable images and Warburg would take them on his travels to lecture on visual motifs.
Later, in the 1930's, another art historian Erwin Panofsky formulated concepts of iconology, which "concerns itself with the subject matter or meaning of works of art as opposed to their form." In his essay, "Iconography and Iconology: An Introduction to the Study of Renaissance Art (pp.26-41)," he sets up three classifications of interpretation:
Panofsky describes the usefulness best when he states that:
"Iconological interpretation, finally, requires something more than a familiarity with the specific themes or concepts as transmitted through literary sources. When we wish to get hold of those basic principles which underlie the choice and presentation of motifs, as well as the production and interpretation of images, stories and allegories, and which give meaning even to the formal arrangements and technical procedures employed, we cannot hope to find an individual text which would fit those basic principles as John 13:21 fits the iconography of the Last Supper. To grasp these principles we need a mental faculty which I cannot describe better than by the rather discredited term synthetic intuition, and which may be better developed in a talented layman than in an erudite scholar. (p.38)"
(Perhaps Panofsky supported the diverse skills of an information professional within the highly researched fields and its scholars.)
Panofsky did not see these three categories of meaning as independent, but rather a way to "refer to the aspects of the work of art as a whole," and when applied to access through an electronic database, sets up cross referencing that Guimier-Sorbets envisioned and outlined in 1993 when she spoke of querying texts through natural language:
"Extending public use will necessarily entail different usages because, from the same specialized information, the system must not only meet the needs of the researchers and help educate students and pupils but must also meet the expectations of a wider public. (p.214)"
Though Panofsky's and Guimier-Sorbet's end goals differ due to their perception of user and the changing attitudes of collection sharing, the depth of access they desire sound remarkably similar, even with the half century division. What Panofsky states as near necessity for understanding and analyzing art, Guimier-Sorbets begins to realize in systems robust enough to offer "linguistic analysis followed by a statistical processing designed to calculate the information weight in relation to all the texts in the base, and to each concept identified and indexed. (p.215)" However, this comparison runs ahead of the evolving philosophy of image retrieval. The five decades which separate these two philosophies shows an interesting progression of thought and technology.
In 1988 Dierdre Stam lectured on the history of the computerized cataloging of art objects, categorizing three decades from the late1960's through the 1980's. First, the unity of the 1960's, which was "characterized by a vision of large-scale, multi-institutional systems created through concerted effort...mentioning libraries as allies (p.8)". Then, the multiplicity of the 1970's, documented by the setup of different cataloging systems for specific purposes, which was drastically affected by the limitations of software. Inclusive in this time period was the arrival of on-line catalogs within art museums such as DARIS at the Detroit Institute of Art which eventually served all of Michigan by 1980. Finally, the harmonization of the 1980's, which marked the "integration of systems across departments and functions within individual institutions, the construction of authority lists with thesauri, and the identification of a core of useful data elements. (p.13)" However, the approach to the timespan was to illustrate the three decade effort to develop a code for the description and cataloging of art collections without success. Yet, by 1991, Jocelyn Small writes some defendable design principles for databases used to retrieve images verbally and lists principle seven as, "No codes...ever. (p.57)" Small uses the classification system of ICONOCLASS to support her argument against codes,
"ICONOCLASS is based on what I call the Roget Thesaurus Approach to the World. Take everything, absolutely everything you can think of, then divide it into one hierarchical and interlocking set of broad categories...add codes so that additions are difficult to fit in and when enough accumulate-and there are nine volumes to ICONOCLASS-impossible to remember and practically unbreakable. (p.57)"
Small also faults ICONOCLASS with consistently omitting the information the iconographer really wants to know due to its overly specific descriptions, yet sometimes inaccurate or misleading data of the art work's subject and contained objects.
The remainder of the design principles constructed by Small follow (p.60):
1. "Preserve the Mess". Do not make your datum more accurate than it is.
2. Information should be reduced to its smallest unit. or least common denominator.
3. Never will any person or project, no matter how knowledgeable and experienced, be able to put together a list of words that will not need to be changed...continually .
4. No controlled vocabulary should be produced in the absence of actual usage.
5. The amount of effort to record a particular piece of information must be weighed against the amount of usefulness returned from that piece of information.
Corollary 1: Utility always takes precedence over philosophy.
Corollary 2: Not all information useful to a scholar is worth recording.
6. It is easier to catalog whole groups of entities than to remember which ones are the right ones.
7. No codes...ever.
8. Everything, and especially free text, should be fully and Boolean searchable.
9. You should always buy a program more powerful than you anticipate you will need, since inevitably you will not have anticipated everything you will need, for which compare Principle Three.
10. The official rule for relational databases that no piece of information should be repeated exists only to be broken. If using the database is easier with information repeated, then do so. Storage is the least of problems.
Corollary: Multiple and overlapping points of access allow for better retrieval.
Now, in the 1990's, those concerned with image databases are particularly concerned with the intellectual problems of cataloging as well as systems servicing a wider user need. In 1988, Michael Krause addressed the hard and soft aspects of indexing describing hard indexing as being "concerned with description, what an indexer can see in the frame, what a picture is of, [while] soft indexing relates to meaning, what an image is about, the personal reactions and feelings stimulated (p.73)." Now, we find ourselves back to Panofsky's essay with a more workable vocabulary for librarians involved in cataloging images. Yet "librarians sometimes say they stop at description, that they cannot go any further...showing that description often includes some comment on meaning (Krause, p.74)." This avenue begins the complex and serious intellectual questions that must be answered before erecting an image database. However, the ability to apply these ideas together (that is larger access with complex retrieval points) could not have been realized without technology's advancement of systems. But, to the intellectual structure of retrieval, Krause offers pertinent questions (through example) for us to consider before erecting an image collection (pp.74-81):
"Here are four terms, all with very similar definitions and all in common use, yet each has a different meaning in the public perception: freedom fighter, guerrilla, partisan, terrorist...when an indexer chooses a heading , he or she makes a moral and political decision-even if the same heading is used for all four categories.
Pictures prompt an immediate reaction and users can see an indexer's choice of heading at once, agreeing or disagreeing on the basis of a glance , whereas books have less instant impact.
If a photograph was taken using a filter or non-standard lens some library users may benefit from access to this information; therefore there should be the facility to retrieve images using certain techniques. (Maybe inclusive in this field could be image manipulation)
The most interesting point Krause makes concerns the indexing of buildings. He states that an image of a building may contain more information than is indexed. Take, for example, a pre-World War II theatre that illustrates the architecture of the time but also contained within the picture is its current use at time of photo capture, a bingo hall; or a non-conformist chapel in the Victorian neo-classical construction that was converted to an Italian restaurant generations later.
Due to copyright issues I have collected images from my own collection to visually illustrate these points:
This image is a building along the C & O Canal on the Virginia side. Historically, it was a hotel for travellers along this waterway. Today it houses the historical artifacts of the canal, its workers, and visitors.
Behind this handsome man you see the building with a variety of signage: Blakesburg Oil Co., Mobiloil, and the symbol of Pegasus. When this picture was taken, the building was an independently owned gas station not affiliated with Mobil Oil. The original use (which dictated the architecture) was an oil company that serviced the community with its needed coal and oil at the height of this small Iowa farming town. The population at that time (post war 1940's) was around 2,500. Today it struggles to hold on to 500 inhabitants. At one time after this photo, it was a well-loved restaurant.
As a photographer, I'm constantly interested in the evolution and renovation of buildings and communities. The stable images possess little interest to me. The three images that follow also illustrate Krause's points as well as the flux that impacts cataloging/indexing:
You probably recognize this building on Main Street here in town. Daily, I watched the care the workers took to preserve the remaining reliefs.
Should the evolving uses/changes of the buildings also be indexed so users could visually support research in the economic, religious, and cultural changes of that society? Should more importance be placed on this aspect of indexing?
With so many issues to address, we can see why developers of image collections have specialized software made to meet their projected philosophies. But with the onset of customized software and the lack of truly constructed cross-platforms in systems, the future of image access uncannily resembles the historical change from local card catalogs to integrated OPAC's. The next question seems inevitable, Should we focus more on developing systems that accommodate our access desires, instead of constantly plugging visual information into the lacking systems that exist? I have sinced realized the remarkable progress made since Saxl first constructed those wooden frames, thus prompting more support for all ideas and visions concerning broader, deeper access.