Skip to main content
DeutschDE
European Commission logo print header
Inhalt archiviert am 2022-12-23

Pattern recognition using parallel systems based fast Fourier transform on finite noncommutative groups and Lie groups

Ziel



One of the main problems of informatics is to clarify how a person's eye and brain recognise the objects of the real world. All human practice shows that the eye and brain solve this problem successfully and recognise objects at different distances and for different extents of blurring.

A perspective approach to the solution of this problem is based on the application of algebraic invariants and abstract Fourier transforms on Lie groups enabling representative sets of features of patterns to be recognised that are invariant with respect to a wide range of image geometrical distortions. But the great computational complexity of this approach blocks its digital realisation. It is proposed that this problem is solved using the theory of fast Fourier transform (FFT) on noncommutative finite groups and Lie groups. Unlike classical FFT, the theory of which is now complete, the total theory of FFT on arbitrary noncommutative Lie groups is at the stage of active development. The main points of FFT theory on noncommutative Lie groups of 1-, 2- and 3-dimensional transforms have been proved by the authors of the project and can be applied to the elaboration of spectral-correlation pattern recognition fast algorithms both on sequential and on parallel computers. Experiments conducted on personal computers have confirmed the high efficiency of the developed algorithms and the first joint work with PARSYTEC, directed towards creating an automated system of hand-written text recognition for bank documentation reading, showed in principle the possibility of effectively realising the suggested approach

The project aim is to create hardware and software transputer complex for hand-written texts recognition with expertness and artificial intelligence elements on the base of design of imitational simulation and comparative analysis of pattern recognition fast algorithms for images subjected to spatial-geometrical distortions.

Programm/Programme

Thema/Themen

Aufforderung zur Vorschlagseinreichung

Data not available

Finanzierungsplan

Data not available

Koordinator

Parsytec Eastern Europe Parallele Computer Gmbh
EU-Beitrag
Keine Daten
Adresse
Bernsdorfer Strasse 210-212
09126 Chemnitz
Deutschland

Auf der Karte ansehen
Gesamtkosten
Keine Daten

Beteiligte (2)

Diese Seite teilen

Herunterladen

Letzte Aktualisierung: 19 September 1996

Mein Booklet

Permalink: https://cordis.europa.eu/project/id/INTAS-94-0708/de

European Union, 2024