Dear Natasha Goldman, dear H-Arthist,

As the question of database structures for (digital) images concerns us all
more or less, I would like to contribute my modest knowledge on that
subject. This answer does not deal with one special program.

In this e-mail I'll provide information on six areas:
1. Possible database structures
2. The role of IDs and norm-data
3. Image quality and storage
4. The nature of available services (image sellers, content brokers...)
5. Where to get information, how others do it
6. Some words on money and ideology

1. POSSIBLE DATABASE STRUCTURES
First, the problem of database structure should be treated independent from
the system used, or even more important from the system you would like to
use.
Structure is a property of the data. Therefore it should not be distorted by
some kind of tool, which is not adaptable to the data.

Maybe it is too exaggerated, but typically there are two kinds of database
structure dealing with (digital) images. The first is what I would like to
call the phonebook-structure. The second kind is the complex research
database.
The choice what to use should be dependent primarily on the data you already
have or you would like to collect.

The phonebook structure is usually very flat, which means that there are
very few hypotactical relations inside the data. As a consequence the data
itself is often very redundant.
The phonebook-structure is very similar to a card catalogue in an old
fashioned library or to the stickers you will find on slides in an old
fashioned visual resource collection. Databases using that kind of structure
usually serve as a finding aid for the patrons of a slide library or
something similar.
A great percentage of phonebook-applications use very simple, often off the
shelf database products and are very happy with that. Concerning the
definition of the structure, standards are used widely, the VRA-Core and the
Dublin-Core being the most well known. Recently more sophisticated systems
have been developed, which still depend on the very flat
phonebook-structure. They serve different purposes like multi collection
search, very comfortable image viewing or building presentations for a
lecture...
More than that, the retrieval tools you can use on your data get more
sophisticated every day.

The phonebook structure is most likely your solution, if you are building a
visual resource collection from scratch and your main purpose is providing
images to your patrons.
It is definitely not your solution if you have data already or would like to
work with data, which is more complex than the phonebook-structure.

In that case the complex research database is your solution: Many of the
flat structures describe the image and the object depicted on the image as
one entity. This habit is similar to the label on an old fashioned slide,
where you could read "Bernini, Apollo chasing Daphne, 17th Century".
For a phonebook-application this is sufficient and widely used (for e.g.
at prometheus-bildarchiv).
Outside the phonebook this practice does not make sense.
First, the image is not equal to the object depicted. Bernini is not the
photographer. The slide is not dated in the 17th century.

Moreover it is impossible to describe more than one object depicted in one
image in such a flat structure: Consider a slide with two sculptures, one
ancient, the other by Bernini. How would you enter the date information in a
single field without dating Bernini¹s sculpture to antiquity and without
repeating information entered in other fields?

Many collections dealing with (digital) images already have a lot of data.
The structure of this data is as diverse as human cognition. Adapting this
data to any standard of the day is in any case a minimal solution. Like
human cognition, the data cannot be standardized. If possible, the structure
inside the data should be kept. The downsizing of information to any
standard of the day can be automated in most cases for secondary purposes.
What seems complex at first is much better maintainable with a complex
system than with any "simple" solution. Even the most simple field
definitions can be distorted by wrong entries.

Dates, attribution of authorship, even the identification of the work itself
are subject to scientific opinion. A lot of the data collected for centuries
already reflects this, by including citations and other useful information.
This good practice should apply for electronic tools as well. Let me explain
this in a real world example:
At the Munich Glyptothek there are separate card catalogues concerning
object information, photographic documentation, bibliography and provenance
information. Each of them dates back to 1850 or even earlier. The object
information contains notes on the personal opinions of the Glyptothek staff.
This information is very precious, as it contains opinions which were never
published.
The bibliography does not only contain the references to the literature, but
even the date and style the author attributes to the specific object. The
catalogue on photographic documentation is an important source for the
historian of photography as well as for the archeologist, as it contains
information on the photographers.
All these information has been incorporated into the electronic archive. A
large percentage of this information would have been lost using a flat
standard, like for e.g. the Dublin Core. The originators of opinion would
now be untraceable.

The solution for the complex database structure is the semantic web, in
which all dimensions of data can be normalized as desired.
It is not limited to a specific structure or rules, which you will find most
likely in any relational or object oriented application off the shelf. The
connection of the semantic web to a flat structure is very easy, as the flat
data can be extracted from the complex data on the fly. The advantage is
that not only today¹s phonebooks can be provided with your information, but
also the systems of the future, which are able to transmit more complex
information.

Guidelines for a semantic web data structure are provided by the
ICOM/CIDOC-CRM initiative. (see http://cidoc.ics.forth.gr/ )
The system used at the Munich Glyptothek is DYABOLA, an application which is
capable of addressing all the subtle complexities appearing in research
databases, museum inventories, excavation databases and library catalogues.
It is available free of charge for interesting projects. (see
www.dyabola.de)
The structure can be adapted to the data by the researcher without
programming. The downside is that the researcher has to understand the
structure.
This however is in the nature of the application, as structure is a property
of the data. It isn't the business of some specialist in the information
science department. It's the business of the researcher.

2. THE ROLE OF IDs AND NORM-DATA
IDs are very important. Every Item in your collection and every other
normalized dimension of your data (person, location...) should be provided
with some kind of unique identifier. This is very important, because
otherwise nobody is able to cite or refer to your data.
Still there are large visual resource collections containing 700'000 photos
without such IDs. As a consequence you will end up searching photographs for
hours, as they change their location inside the collection frequently over
the decades. In an electronic environment missing IDs generate fuzzy
results, which is as annoying as running through the collection for several
hours.
Providing IDs is very simple. Most likely every item in your database has a
record number or something similar. Make sure that this number is visible to
your patrons. A combination of the record number with some kind of URL or
DOI (digital object identifier) will make your records referable from
anywhere in the world. The whole identification should be legible very easy,
as the world comprises sheets of paper also, not only machine-read
anchor-tags in some sort of hypertext.

Norm data like artist names, geographic locations, and so on are useful,
especially when communicating your data to patrons via some kind of broker
(see below).
However there are two issues to be cautious. First there is not and
hopefully won't be any kind of brazilianyte "central services".
For artist names you could choose the already established ULAN or AKL.
Alternatives are the PKNL (prometheus), which was released last week or the
this and that list, which will be issued next year. Incorporating one or all
of them into your data is an effort which has to be considered. It can be
automated to some extent, but proper control means some manual work. It can
be useful, as normalizing artist names is serious intellectual work and it
does not make sense to do the same work all over again.
Second caution: All the norm-data will be useful to identify the well known
artists, locations, etcŠ. When comes to the long shallow tail of the data,
the norm-data won't be very useful. In a mathematical sense your data is not
normal very likely.
Consider "The guy from Naples who paints sheep very similar to, but in my
opinion is not 'the guy from Naples who paints sheep (ULAN-No. this and
that)'". In that case it does not make sense to send an e-mail to the AKL or
ULAN. Maybe you change your mind next week and the name you have provided
does not make sense any more. Working with renaissance drawings or baroque
painting this can be very important, as you will end up with up to 70%
anonymous or loosely attributed material.
Adding the ULAN-Number to an attribution does not reinforce the attribution.
To be sure, adding the ULAN-dataset of Piranesi to a drawing does not make
it more piranesian.
In the end data which is annotated with norm data can be brilliant or crap.
Data can be brilliant too, not using norm data at all.

3. IMAGE QUALITY AND STORAGE
Go for the best quality you can get. Display technology will improve to an
enormous extent in the next years. Today¹s state of the art displays for
laptops have a resolution of 1920 x 1200 pixels. In next generation video
this resolution will be 16(!) times higher.

If image sellers provide you with high resolution images they deliver 3000 x
2000 pixel, the scanning standard of a regular 36 x 24 mm slide. To be
precise this is only "high resolution" in comparison to a 1980s home
computer.
Compared to a "Lichtdruck" by Brunn-Bruckmann from 1905 this resolution is
lousy. Printed in that quality, the hi-res scan won't be any larger than a
few square inches. The hi-res images are definitely not the material
Wölfflin and Warburg would have chosen.

Today high quality printing means 550 to 600 dots per inch per color with up
to 10 colors (forget RGB). That means what you can see on a regular book
page is light-years from the best beamers available.

With an eye to the future scans should contain as much information as
possible. Storage cost is not a problem any more.

The images can be archived in the original resolution and either be
downsized to the specific use on the fly or be held separately in a lower
resolution. It may be useful for that application to make use of a separate
image server, which basically functions like a file-server and is
independent from the database system containing the meta-data. On such
servers the images are stored according to their ID in the database. The
server delivers the image in the preferred form on the fly (resized,
filtered ...). To separate the database from the image server is useful, as
image databases providing both are a mixture of average solutions on both
sides more often than not. If possible go for a media server which can
handle not only images but all kinds of media.

Make sure that not only the collection database has access to the images on
the server. The image server can also be used as a source for student's or
researcher's projects, which use other database systems or simpler solutions
like static html web pages. With a solution like this you could manage an
inventory of all relevant images at your institution and in the same time
others would be free to experiment, all while using the same images.
With that solution, lecture systems (power point, acrobat or proprietary) do
not have to be standardized.

4. THE NATURE OF AVAILABLE SERVICES
Image sellers are companies which deliver slides and digital images often
including phonebook-style metadata (MARC-format, VRA-CORE...). The most well
known in the U.S. are Saskia and Davis. There is no need to mention, that
the material provided by these companies is much better than most of the
copy-stand photography taken from books.

Content brokers are services, which communicate your images to a wider
audience and/or deliver images from other sources (via a collective
interface). Most of the time, a flat data standard is used in the operation.
In that case complex data from your database will be simplified. Some
information will be lost. On the other hand, your information reaches a
wider audience.
Being a member of some sort of brokering service will also enrich your
collection, following the rule give 500 images and get 100'000. Subscribing
to a brokering service, should depend on your purpose, as data quality is
differing. Some providers like AMICO specialize on quality images and data
from museums and other collections. (see www.amico.org )
Other services like Prometheus-Bildarchiv in Germany also focus on
connecting the existing slide libraries of the member institutions, which
are mostly university departments. Here the images are often copy-stand
photography, but nonetheless the service is more useful than any single
slide library.

As with any other system there will be no exclusive solution. In the recent
15 years there have been many attempts to establish some kind of "central
services". As with standards, new services are established frequently,
CollectionConnection mentioned by Geert Souvereyns being the latest example.
In the end there won't be one single "central service". Even google catches
only a small part of the known web.

5. WHERE TO GET INFORMATION, HOW OTHERS DO IT
The problem of how to collect, maintain and provide (digital) images is one
of the most important tasks of art research, especially in an institutional
framework. It is well known for a long time, that collaboration between
institutions and individuals is crucial. However as in any other field of
activity it is very difficult to get bias-free information. Like in politics
there are different lobbies. Therefore when building a new database, it is
useful to consult all of them.

The group you belong to most likely according to your question is the art
information professional, which includes art librarians and visual resource
curators. Your duty is conservation and providing patrons (i.e. researchers)
with material.
Most likely you will focus on developing standards of good practice and
reduce work by symbiotic collaboration with others.

The second group is system providers, i.e. people and companies producing
database systems and other applications. They will most likely tell you that
their way to do it is the only one and that all other systems are crap. Be
aware of the open-source guy at the coffee-machine. He will be the most
annoying, as his system is not a tool but a religion. I personally use a lot
of open source products too. Open source software is a very good thing. But
in the end it is a tool just like the bone or the space station. Remember
that the structure is in the data. The system has nothing to do with it, if
it is capable to reflect the structure. (by the way: open source has nothing
to do with open access)

The third group is service providers, i.e. image sellers and content
brokers. They are a very valuable source of information, as they work with a
lot of data. Usually they can tell you more about structure than the
theoretic guy from the information science department. On the other hand
they use standards which might be too strict for a complex application.

The fourth group is the most important. It is the group of the user, i.e.
the researcher. This is the group where you and I belong to personally, not
as a member of an institution. As a lobby we are underrepresented. Therefore
our demands are the greatest. What we want is unlimited bandwidth, as George
Lucas once put it. Our duty is to take all the tools and all the information
and play a symphony.
We want to know all the complexities of any data. Search and browse it with
different tools. Get the best image quality. Rip, mix, and burn it. Remix
it. Sample it. Scratch it. And data mine it.
We do that in order to extract new information from the data, to produce new
data, to verfiy the data, to falsify the data and find out unknown
properties of the structure.
You can be sure that our way does not comply with any standard. Most of the
predefined standards will fall in the progress of our research. As William
J.T. Mitchell pointed out recently, art research is science. And science,
according to Paul Feyerabend, is basically an anarchic enterprise.
Therefore the data should be as open as possible.

Below there are some references, belonging to the first group, that of the
art information professional.
I won't contribute the addresses of the system and service providers, as
this field is constantly growing, and any listing would be idiosyncratic.
Most of them advertise. You will find out, that some of the best providers
are subject to damnatio memoriae among some other groups of people. The
reason for this is that they do things, the people you talk to, would like
to sell or put on their own list of honour. Concerning this phenomenon you
should be aware of constellations where the first three groups named above
mix. Institutions should provide bias free access to all information. This
duty does not comply with the development of an exclusive system or one
single service channel.

Advice for the art information professional:
In the U.S. the Visual Resource Association (VRA), a collaboration of visual
resource curators from all over the country, is the most important source of
advice. Here, hundreds of members from small and large Institutions exchange
their Know How and develop guidelines of good practice.
(see http://www.vraweb.org )

A very interesting study by on visual resource collections in the U.S. and
their dealing with digital images is ³Susan Craig: Survey of Current
Practices in Art and Architecture Libraries. in: The Twenty-First Century
Art Librarian. Haworth Information Press (December 1, 2003) p. 91-108²
(see www.amazon.com )

In Europe there is no official profession equivalent to the visual resource
curator as far as I know. Normally the same job is done by librarians or art
historians. However as in the U.S. the maintenance of visual resource
collections is related very closely to the field of art librarianship. It
seems in fact, that the two fields are merging in some extent.
(see final draft 3/10/05 at http://www.uflib.ufl.edu/afa/pdc/coredata.htm)
A lot of countries in Europe have an association equivalent to the Art
Libraries Society of North America (ARLIS/NA).
(see www.arlisna.org )
In Germany this is the "Arbeitsgemeinschaft der Kunst- und
Museumsbibliotheken", which provides a comprehensive List of similar
Institutions.
(see http://www.akmb.de/web/html/links/fachverbausland.html ).

It is important however, that the art librarians¹ point of view is
characterized by the handling of library material. Books and articles
usually have a title, an author, a date of publication and a location of
publication. Therefore it is possible to standardize the data to some
greater extent.
Works of art, which are the real subject of most (digital) image databases,
do not have these clear cut properties usually. Therefore you should not
believe everything they tell you about standards.

6. SOME WORDS ON MONEY AND IDEOLOGY
The adaptation of a database system to your own needs is an intellectual
piece of work. It does not matter if the system is open source or some other
product - somebody has to do the work.
If your purpose is very similar to someone others, chances are good, that
you will get a working system at a very low expense. If the adaptation is
more complex and individual, it will cost some time and/or money no matter
whether the system used is a commercial product or a piece of open source
software - somebody has to do it.
With open source, we have to be aware, that our purposes are in no way as
widespread as bittorrent or mozilla firefox. Therefore chances are that
there are very few programmers working just for fun on your special problem
of the day. In the end you will end up paying for the work.

In fact there are many models which can be successful. You can buy a
standard application off the shelf, use open source software or even write
your own application. Paying some commercial provider can be much cheaper
than hiring a self-appointed specialist, who is going to invent the wheel
again.

Asking people with a lot of experience in the field will certainly cost some
money. On the other hand, it might keep you from making mistakes others have
made, which will most certainly cost a lot more.

The problem of "dependence on exclusive solutions" is a thing which has in
no way to do with the database system or the licensing model used to spread
it.
You should insist, however, that the data is exportable from your system
without the help of a programmer.
Given that, you are save if the company goes bust or if your three open
source programmers are unreachable, because they got a well paid job at the
NSA.

For a funny account on open source and ideology in Europe see Bruce
Sterling in WIRED 11.09
( http://www.wired.com/wired/archive/11.09/view.html?pg=4 )

With kind regards,

Maximilian Schich.
maximilianschich.nu
www.schich.nu

(The author is art historian; his continuous professional engagement in
art information dates back to 1996)