A Recognition System of Multi-voice Music Staff

Dengsheng Zhang     Anqing Normal Institute, Anqing, 246001, China
                           Songde Ma                National Key Lab of Pattern Recognition, Beijing, 100080, China

ABSTRACT

This paper presents a complete system to interpret original musical image to computer readable musical text. It first segments the whole image into staves which are then further segmented into individual objects. The objects are then recognized by extracting out their uneven density features and using knowledge-based rules. The recognized objects are then registered, encoded to semantic code and restored to original image. The rule set used can easily be redesigned or modified according to the syntax that governs the meaning and placement of symbols in a particular manuscript class, showing its strong sensitivity and adaptability. Recognition rate for discrete notations reached 98%, and for complete staves is 92%.

Key words:Staff, Sytax rule, Semantic recognition, Uneven density feature, Projection.