Accessing the visual heritage: metadata construction at the Science & Society Picture Library
| DOI | https://doi.org/10.1108/eb040645 |
| Pages | 58-64 |
| Date | 01 March 1997 |
| Published date | 01 March 1997 |
| Author | Angela Murphy |
| Subject Matter | Information & knowledge management |
Accessing the visual
heritage: metadata
construction at the
Science & Society
Picture Library
by Angela Murphy, Science & Society
Picture Library, Science Museum,
Lon-
don,
and Peter Enser, School of
Information Management, University of
Brighton
The Science
&
Society Picture Library —
responsible for a very large archive
of still
images drawn from many different collections
—
is presently confronting the challenge of
developing an integrated cataloguing and
indexing strategy
by
which metadata
construction can
proceed,
and which will
provide potential users with effective and
standardised subject access
to
the many
components
of its
holdings.
Introduction
Technological advances in the capture, compres-
sion, manipulation, storage and transmission of
electronic images have excited the attention of a
rapidly growing community whose responsibilities
embrace curatorial, entrepreneurial or research
roles with respect to our visual heritage. The
characteristics of these technological developments
have been reported in depth(1,2) and need not detain
us here. Our attention must be drawn, however, to
the fact that, in combination, these developments
have operated with accelerating force on our
ability to gain physical access to visual heritage
without offering any comparably effective means
for gaining logical or subject-based access to that
rich vein of pictorial knowledge (3,4).
The retrieval of appropriate pictorial material in
response to client need has long exercised the
guardians of our visual archive. It has taken image
digitisation to bring this particular information
retrieval problem to prominence in the research
agenda, however.
As a data structure, the digital image offers entic-
ing processing opportunities for the research
communities in computer science and artificial
intelligence. Operating on the morphology of the
underlying pixel structure, similarity analyses can
be conducted which seek to recover from an image
database those images which are sufficiently alike
a target image (the 'query') to be considered
worthy of presentation to the client. Such content-
based
image
retrieval uses quantifiable attributes
such as colour, texture and geometry (i.e., shape,
spatial distribution or — in the case of moving
imagery — spatio-temporal distribution of regions
or segments within the image) as distinct from a
traditional, semantic information retrieval opera-
tion which operates on the textual attributes
available in the image metadata.
The new approaches to content-based image
retrieval have borne fruit in a number of admirable
applications(5-10). Typically, such prototype sys-
tems have used small and highly constrained test
collections of query images, however. Although it
is likely that new digital image capabilities will
bring forward clients with different visual informa-
tion needs, the experimental content-based systems
operate at a very considerable remove from the
currently expressed needs of users of visual herit-
age collections, examples of which have been
reported elsewhere (3,4,11,l2).
Within archival image collections a heavy depend-
ency continues to be exhibited on a traditional
visual information retrieval model in which the
query is verbalised by the client, refined by a
picture researcher and, hopefully, resolved as a
simple text matching process between the (medi-
ated) terms in the query and those present in the
image metadata in the form of
title,
caption and
keywords. A successful outcome is likely to need,
in addition, the application of the picture research-
er's knowledge about the collection and the subject
domain, together with his/her world knowledge
and some browsing time. Significant reservations
have been expressed about a visual information
retrieval process which operates in this way,"
dependant upon a prior translation into linguistic
terms of the content or meaning of an image(3).
Nevertheless, certain image retrieval needs (those
which, in demanding a uniquely defined object,
58—VINE
107
To continue reading
Request your trial