基于改进特征提取方法的五线谱识别  

Recognition of Musical Notation Based on Improved Feature Extraction Algorithm

作　　者：陈琢 吴亚联[1] 何婕[1] CHEN Zhuo;WU Ya-lian;HE Jie(College of Information Engineering,Xiangtan University,Xiangtan 411100,China)

机构地区：[1]湘潭大学信息工程学院，湖南 湘潭 411100

出　　处：《软件导刊》2019年第6期125-129,134,共6页Software Guide

基　　金：国家自然科学基金项目(61379115)

摘　　要：针对传统五线谱识别方法存在谱线过删和漏删的缺点,以及现有音符特征提取方法与谱线删除相互制约的问题,提出一种改进的、无需删除谱线的特征提取方法。在图像预处理阶段保留谱线,将音符与谱线同时投影,结合音符符杆垂直像素数据与音符其它部位像素数据携带的映射特征,对横纵向投影数据进行数理统计分析,得到供音符类型识别的归一化特征值,再利用基准谱线与音符符头的相对位置获取音调信息。实验结果表明,该方法在保证较高识别精度的基础上,进一步提高了识别速率,可以有效识别音符组合形式较复杂的乐谱,对于五线谱识别应用具有重要意义。Concerning the excessive-deleting problem and leaky-deleting problem that existed in spectral delete of traditional recogni-tion of music score,and the mutual restraint between the feature extraction algorithm of notes and the lines delete,a novel feature ex-traction algorithm is proposed,which preserves the spectral lines. We preserved the spectral lines in image preprocessing and project-ed the notes and spectral lines at the same time. Combined with the mapping characteristics carried by the vertical pixel data of notes rod and pixel data of other parts of notes,performing basic mathematical statistical analysis on the horizontal and vertical projection da-ta,the normalized feature values for musical note type recognition are obtained. Then we used the relative location between the stan-dard lines and notes head to obtain the tone information. The results of the simulation illustrate that the proposed algorithm is a feasible way to recognize more complex combinations of notes and improve the recognition rate under the premise of ensuring high recognition accuracy.

关 键 词：五线谱识别 特征提取 映射特征 横纵向投影 归一化特征值 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]