An Optical Notation Recognition System for Printed Music based on template matching and high level reasoning

In this paper we present a laboratory prototype that has been developed at the Institute for Language and Speech Processing (ILSP) for Optical Recognition of Printed Music in die framework of the MUSTUTOR project funded by the EC. The algorithms implemented proceed in three different stages and perform symbol recognition of the printed music image. The system also includes image scanning and file handling capabilities, whereas it can be enhanced by a symbol to audio transcription in order to include the functionality of hearing the recognition results. Two methodologies for recognition have been implemented. The first one keeps the staff lines while the second removes them. Both approaches also comprise an image pre-processing stage as well as a higher level reasoning (post-processing) one, the latter being used for combination of recognition results and application of global music theory knowledge with a set of corrective and enhancing rules. Future enhancements include the use of an expert between the two techniques so as to use a hybrid system for quality enhancement. The system can be used in many applications involving the editing of music scores, the database storage of music scores as well as multimedia applications such as synthesis of existing music works and/or automatic harmonization.