Optical music recognition of piano sheet music

Abstract

Optical music recognition (OMR) is a field in character recognition that focuses on the automatic recognition of sheet music. Musicians use a universal music notation that gives them the ability to express their music through writing instead of performing. The music notation uses modern musical symbols to indicate the rhythms and pitches of a piece of music. In an OMR system a music sheet is usually recognized to output the musical notation to an audible format, for example a MIDI file. However OMR systems have many potential applications, such as the archiving of old music sheets into digital format, automatic transposition of music sheets, and even conversion of a music sheet into braille. In this work an OMR system is designed to be capable of recognizing the basic symbols in a music sheet in order to recognize moderately complex pieces of music, and convert them into a MIDI file. OMR systems are more complex then text recognition since unlike text music representation is two dimensional, containing data representing both the duration and the note pitch. The system is designed to cater for printed music sheets acquired using a mobile or tablet camera. To deal with the complexity of an OMR system the implementation was split into four steps to make it manageable, namely staff line isolation and removal, symbol recognition, pitch recognition and conversion to MIDI. For the symbol recognition stage the COSFIRE filter is used. The use of this filter in an OMR system was as yet unexplored. Thus the feasibility of using this filter for an OMR system is tested by measuring the precision of the COSFIRE filter in symbol detection with an image filled with a large number of symbols in it.