Project PRINCESS
Principles of Dissimilarity-Based Pattern Recognition in Signals, Symbolic Sequences and Images
The project is supported by
INTAS (an EU funded agency supporting cooperation with New Independent States (ex-Soviet Union).
Project duration: April 1, 2005 - March 31, 2007 (24 months)
Consortium
- Czech Technical University in Prague, Faculty of Electrical Engineering, Center for Machine Perception, Prague, Czech Republic, Vaclav Hlavac (coordinator)
- Institute Scientific Council "Cybernetics", Russian Academy of Sciences, Moscow, Russia, Vadim V. Mottl
- Tula State University, Laboratory of data analysis, Tula, Russia, Sergey. D. Dvoenko
- International Research and Training Centre for Information Technologies and Systems, Image Processing and Recognition Department, Kiev, Ukraine, M.I. Schlesinger
- University of Surrey, Centre for Vision, Speech, and Signal Processing, Guildford, United Kingdom, Josef Kittler
- Technische Universität Dresden, Artificial Intelligence Institute, Dresden, Germany, Boris Flach
See PRINCESS team
- photo at the Moscow meeting in April 2005.
- photo at the Prague meeting in September 2006.
- photo at the Dresden meeting in March 2007.
Project Outline
PRINCESS project aims at contributing to theoretical basis and algorithmic technologies of pattern recognition. The emphasis is put to object recognition from signals, symbolic sequences and images. The learning issues, both supervised and unsupervised, constitute important part of the project too.
The traditional (and in a certain sense problematic) approach is to extract informative characteristics from data which represent each object in a
vector feature space. There exists a variety of heuristic ways to evaluate the
dissimilarity between observations - signals, symbolic sequences or images - using an appropriate deformation of their argument axes in the vector space and to form set of two argument functions each of which possess all the properties of a metric. The projects aims also to express and utilize structure observations to enhance recognition capabilities.
A structural approach implies inferring the final decision from descriptions of parts and expressed relationships among them. The idea of the proposed investigation is to introduce a collection of heuristically chosen metrics used for arranging respective compactness hypotheses to achieve
better predictive properties of the decision rule inferred from a small training set by both statistical and structural techniques.
The project has two objectives:
- To create a general theoretical and methodological framework in the following areas: (1) finding empirically regularities in sets of signals, symbolic sequences and images, (2) dissimilarity-based structural pattern recognition, (3) recognition using two-dimensional context-free languages.
- To transform theoretical knowledge into operational skills. This is going to be performed through conducting pilot applications at University of Surrey, Guildford (human faces), at TU Dresden (utilization of structured stochastic models to 3D reconstruction from 2D images) and at Czech Technical University Prague (structured printed 2D documents of mathematical formulae).
Maintained by V. Hlavac,
hlavac@cmp.felk.cvut.cz
to top