Multimedia Information Retrieval

Massive amounts of digital data are being stored in networked databases
for access via computers or mobile phone networks. Current ICT
technology enables creation of digital data including such modalities as images,
videos, multimedia, animations, graphics, speech, audio and text. A key problem
will be how to organize and index this material to best support retrieval of relevant
data. Not enough man-power is available to index or annotate all the terabytes
of digital material by hand. Automatic and semi-automatic indexing mechanism
should also be implemented. There are limits, however, to which extent it is
sensible to index everything in the database. Content-based multimedia retrieval
is a technology that applies digital signal processing and advanced pattern
recognition techniques to analyse each potential object to determine its relevance.
Applied techniques depend on the media type. For digital images, properties like
texture, colour, and geometric structure primitives are analysed. For videos, movement
of the image parts are tracked as well. For speech signals, automatic speech
recognition, speaker recognition, gender recognition and language recognition
can be used. Automatic detection of various audio events is also feasible today.
A crucial problem is how to bridge the gap between the semantic concepts used
by a user seeking information and the database descriptors automatically derived
from the data. The difficulty lies in the design of reliably computable data features
that describe the data in high-level terms such as semantic concepts. This can be
viewed as a problem of mapping a query language to a data description language.
In the presentation, examples of state-of-the-art data techniques are described that
can be used for retrieving and analysing digital data.