多声部乐谱图象识别系统

A Recognition System of Multi-voice Music Staff

文章提出一个完整的系统，将多声部乐措原文转换成可编辑的乐谱文本。系统首先在乐符检测阶段将图象切分为诸表，然后再将其切分成个别对象。在识别阶段用非均匀密度特征和句法规则将这些对象识别出来，并将其登记，构成语义编码，然后由乐指恢复模块将其恢复成乐谱原文。系统中使用的句法规则可根据乐措类别重新定义或修改，具有很强的灵活性和适应性。系统对离散乐符的识别率为98％，对错段的识别率为92％。

This paper presents a complete system to interpret original musical image to computer readable musical text.Itfirst segments the whole image into staves which are then ftrther 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 particuIar manuscript cIass,showing its strong sensitivity and adaptability. Recogniton rate for discrete notations reached 98 %, and for complete staves is 92%.