STUDENT PROJECT PAGE
Master's project: Thesauri and the Mirror multimedia retrieval model
Querying non-textual data in databases is difficult. In this project, we
attempted to improve the results of content-based retrieval systems with the
addition of explicit knowledge. Mark van
Doorn investigated how and if "thesauri" can
be integrated in the Mirror approach, and maybe also perform some experiments
to test whether thesauri really improve retrieval.
More information
Master's project: Content-based retrieval of music
Two techniques for music retrieval have been described in literature.
Harold Oortwijn implemented the
Musclefish approach to retrieve music based on results from this previous
research. We started with MATLAB procedures to calculate features. Later, we
may integrate these in a module for the Monet extendible database system.
Together with Symbol, a start-up in the
market for radio broadcasting equipment, we evaluated the performance
against a manually classified music collection. The results are encouraging
more research.
Master's project: A software architecture for Mirror and AMIS
In Mirror, we have a pretty complex environment: daemons to calculate
features, clients for retrieval and maintenance, an extensible database
managing the meta information, and a modern multimedia file system to service
the video data. Erik van het Hof
organized this "spaghetti" into a software architecture, with clear
instructions how to write components for this architecture. Java
and CORBA play an important role in this complex environment.
More information
Design project: Applying the GEMINI-algorithm to medical
measurements
Roessingh Research and Development
BV (RRD) is a selfsupporting company focussed on research and development
activities in rehabilitation of the locomotion system. Pieter Kuijpers
investigated the possibility of using database technology to enable the
retrieval of similar patients using the measurements of one patient as
input. This idea may be very beneficial for selection and evaluation of
medical treatment.
Master's project: A neural network for audio retrieval
Traditional speech recognition is not suited for information filtering tasks,
since you have to work with a predefined vocabulary. By indexing phoneme
sequences instead of words, this problem could be overcome.
Hong H. He started realization of a neural
network recognizer for such phoneme sequences in his master's
project.
Master's project: Wavelets and multimedia indexing
The properties of Wavelet Transforms for the indexing of multimedia databases
has been the research topic for the Master's project by
Ivan H.P. Meert. The Discrete Fourier
Transform has been used for indexing time series in previous research. The
first two or three (complex) coefficients can be stored in a spatial indexing
structure to speed up retrieval of time series. Instead of Fourier transforms,
one may be able to use wavelet transforms for the same indexing process. We
expect the wavelet transform to be especially beneficial for the processing of
sub-pattern matching.
Design project: Intelligent Television against Information Overload
"Without information no retrieval!"
This design project by five undergraduate students developed an environment in
which a database is filled with television data from several information
channels. Philips Sound and Vision provided us with a prototype television
that was very suited to be used in this project.
Short design project: Querying Illustra with a Java Client
In this short design project for one student, Remko M. Wiersma developed a Java
wrapper around an Illustra database. This would have made it much easier to
create new client programs on top of Illustra. A case study into course
information on the web has been investigated by Remko.
Last updated: $Id: projects.html,v 1.6 1998/10/20 09:56:01 arjen Exp $
Maintained by: arjen@cs.utwente.nl