Outline of the Artiste Proposal
Summary
The objective of the project is to develop and prove the value of
an integrated art analysis and navigation environment aimed at supporting
the work of professional users in the fine arts
The environment will exploit advanced image content analysis
techniques, distributed hyperlink-based navigation methods, and object
relational database technologies. It will build on existing metadata standards
and indexing schemes
Outline of work:
-
research and develop advanced image content analysis techniques for digitised
works of art;
-
develop techniques to automatically categorise art works using these algorithms;
-
develop metadata representations for image categorisation;
-
develop seamless distributed access to multiple collections;
-
develop distributed content-based navigation methods for art collections;
-
build a robust and scalable integrated environment that incorporates the
above components using an object-relational database;
-
develop a report on the impact on standards detailing augmentations of
Z39.50 with RDF
-
establish business models that give the content owners direct control over
the distributed representation, access and exploitation of digital multimedia
content and metadata;
-
identify how the technical systems developed in the project can be deployed
and exploited in a number of sectors.
Project Justification and Scope
European museums and galleries are rich in cultural treasures but public
access has not reached its full potential. Digital multimedia can
address these issues and expand the accessible collections. However,
there is a lack of systems and techniques to support both professional
and citizen access to these collections.
The ARTISTE project will address professional users in the fine arts
as the primary end-user base. These users provide services for the
ultimate end-user, the citizen. The main target roles are listed
below.
Art researchers, historians and museum curators have a requirement
for dynamic search and retrieval of high-resolution art by image content.
This will substantially increase their efficiency in tasks such as matching
art fragments, detecting and verifying authorship and researching painting
styles and methods. They would like to reduce the time and effort
required in indexing and categorising art works by automating these processes.
Publishers and educational course providers also require access to
high-resolution art, with the ability to search and retrieve art work quickly
using a variety of search terms and the ability to support train-of-thought
analysis.
Innovation
The areas of innovation in this project are as follows:
-
Using image content analysis to automatically extract metadata based on
iconography, painting style etc;
-
Use of high quality images (with data from several spectral bands and shadow
data) for image content analysis of art; Use of distributed metadata using
RDF to build on existing standards;
-
Content-based navigation for art documents separating links from content
and applying links according to context at presentation time;
-
Distributed linking and searching across multiple archives allowing ownership
of data to be retained; and
-
Storage of art images using large (>1TeraByte) multimedia object relational
databases.
Image Content Analysis
The project represents a radical departure in terms of indexing works of
art. Attempts have been made to index the objects and subjects in paintings
using ordered systems, but these rely on expert knowledge of the content
and in-depth experience of the classification system. The ARTISTE approach
will use the power of object-related databases and content-retrieval to
enable indexing to be made dynamically, by non-experts.
The high quality of images in the system, particularly in terms of
colour, will allow comparisons to be made that in the past were based on
information of too low a quality to support the assertions made. Comparisons
made using the shadow data will allow comparison of brush strokes, etc.
Not much research has been carried out worldwide on new algorithms
for style-matching in art. This is probably not a major aim in Artiste
but could be a spin-off if the algorithms made for specific author search
requirements happen to provide data which can be combined with other data
to help classify styles.
In terms of colour matching all previous work has been non-colorimetric,
i.e. varied RGB colour comparisons. The ARTISTE approach will use the unique
collection of CIE colour values from the Vasari/Marc projects to carry
out accurate colour comparisons. Preliminary work at UOS MMRG in
this area shows that one can even begin to match reconstructed spectra
with pure pigment spectra. The ability to put a colour value from say a
colour meter reading of a pure known pigment in as a query to obtain a
search across a collection would to ARTISTE. With similarly accurate collections
such as those in the Uffizi gallery it is also possible to allow authors
to find comparisons between collections, which were just not possible with
non-calibrated images. NGL's fading data can also be used to provide clues
about matching to colours before 100 years of light fading for example.
This could help in cases such as textile/fabric sample matching.
Based on experience from working with art historians and art gallery
professionals MMRG is aware of requirements in retrieval of art,
rather than classification. In other words ARTISTE would aim to give searchers
tools which hint at links due to say colour or brush-stroke texture rather
than saying "this is the automatically classified data". Also by concentrating
on specific examples such as fabric finding, pigment finding, specific
shapes etc., we are producing new research results rather than trying to
satisfy a huge array of users, which has been shown to be the weakest approach
in content-based retrieval.
Indexing and Metadata
The ARTISTE project will build on and exploit the indexing scheme proposed
by the AQUARELLE consortia. The ARTISTE project solution will have
a core component that is compatible with existing standards such as Z39.50.
The solution will make use of emerging technical standards XML, RDF and
X-Link to extend existing library standards to a more dynamic and flexible
metadata system. The ARTISTE project will actively track and make
use of existing terminology resources such as the Getty "Art and Architecture
Thesaurus" (AAT) and the "Union List of Artist Names" (ULAN).
Integrated Art Collections
ARTISTE will integrate art collections while allowing the owners of each
collection to maintain ownership and control of their data. This will be
achieved by virtually integrating the collections using the concept of
distributed linking. The distributed linking will add links to content
(both text and images) at presentation time. This will enable a user to
add links to content that they do not own or have write access to.
Distributed links will ease the management of links by separating them
from the content. This means that new links can be applied to an existing
resource without modification to that resource (for example when a new
image content analysis algorithm has been performed on the data set). In
addition, different sets of links can be applied depending on the user
viewing the resource.
Object Relational Databases for Storage of Art Images
The proposed architecture will allow multiple distributed databases to
be integrated, removing the need for centralised repositories. However,
as volumes are expected to exceed a terabyte, ARTISTE will use scalable
object-relational database technology to manage these large data volumes.