Feature extraction for the recognition of musical primitives

Optical Music Recognition (OMR) aims at reading automatically scanned music scores, in order to convert them into an electronic format, such as a MIDI file. Although some commercial products are already available, music score recognition is not error free.
In order to improve the reliability, we propose new analysis methods based on the fuzzy modeling of symbol classes and of musical rules. A software has already been developed for monophonic music processing, and leads to good recognition rates compared to the performances of SmartScore.
The symbol analysis is basically based on template matching. Correlation scores are used to define fuzzy classes and possibility degrees. The purpose of the project is to find new symbol analysis features that could improve the reliability of the symbol analysis: for example, geometrical features, moments, etc. These new features must be integrated in the fuzzy model of symbol classes. Tests on a large database will be conducted, in order to evaluate the contribution of the added features.

Keywords:

Image processing, pattern recognition, fuzzy modeling

Requirements:

Basic knowledge in signal Processing, Matlab and/or C languages

References:

OMR research project

Contacts:

Florence ROSSANT