乐谱识别关键技术问题及其解决方案  

Key Issues and Solutions of Optical Music Recognition

作　　者：刘晓翔[1] 周密[1] 陈双平[1] 

机构地区：[1]暨南大学电气信息学院,广东珠海519070

出　　处：《计算机仿真》2015年第7期253-258,385,共7页Computer Simulation

基　　金：国家自然科学基金(61201458)

摘　　要：分析讨论了乐谱识别关键技术的研究现状及其存在的问题,针对多声部乐谱识别的关键技术提出了相应的解决方案。1提出一种基于投影和相关计算的谱线检测方法,能够克服变形、相交和粘连的干扰;提出一种基于游程块邻接图的谱线删除方法,有效减少图元的过删现象。2提出一种基于结构特征约束的音符基元抽取和重组模型,使系统在复杂环境下的适应性与可靠性得到提高。3提出一种基于声部划分的多声部乐谱识别校验方法,将识别反馈机制从单声部乐谱延拓至多声部乐谱。在此基础上开发了乐谱识别原型系统IOMRS,与商业化乐谱识别软件进行对比评测的结果表明,IOMRS的音符识别率高、稳定性强,尤其在识别复杂版面的多声部乐谱方面具有优势。This paper discusses the research status and problems of optical music recognition（ OMR）,and studies several key issues on polyphonic score recognition and proposes the solutions.（ 1） An approach for detecting staff lines is proposed based on correlation computing and horizontal project,which solves the problems of distortions,interconnections and adhesions; and an approach for removing staff lines is proposed based on run-length block adjacency graph,which reduces the risk of ＂over-removal＂.（ 2） A model for primitive extraction and reconstruction is designed based on structural feature constraint,which improves the flexibility and reliability in complex data environment.（ 3） An approach is proposed based on voice assignment for checking and rectifying recognition results,which expands the verification mechanism from monophonic scores to polyphonic scores. Based on our approaches,we develop an OMR prototype system called IOMRS,and test it against popular commercial OMR products. The testing results show that the IOMRS has good performance with high recognition rate and strong stability,especially for the case of complicated polyphonic scores.

关 键 词：乐谱识别 多声部乐谱 谱线检测与删除 音符识别 识别校验 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]