摘要
提出基于图段拓扑关系的谱线删除方法,以避免谱线过删除现象;提出双向游程编码结合使用的符干分割方法,克服了现有方法对复杂音符适应性差、分割结果不完整等缺陷;提出音符先验知识引导下的符头切割与检测算法,以解决粘连符头的切分问题;提出基于块状体分割和特征检测的符梁分割算法,设计了适用于乐谱版面的文字和线条提取算法。该方法应用在乐谱识别系统中分割乐符具有良好的性能,尤其对乐谱内容复杂、乐符排列密集等情况有较强适应能力。
Firstly, proposed a staff lines removal approach based on analyzing topological relationship among run-length segments, as a result, staff lines over-removal could avoid. Secondly, proposed a stems segmentation approach, using both vertical and horizontal run-length coding. It could overcome the weaknesses of current methods such as the fragmentary results and poor adaptability to the complex notes. Thirdly, presented an approach for note heads segmentation, employing the prior knowledge of music notes to split and separate touched note heads. Lastly, designed a block-based approach for beam segmentation and an applied approach for text and curves extraction. Implemented the approaches in one optical music recognition system. It shows good performance and powerful adaptability, especially for the situation of intensive layout of music symbols, or the case of complicated music score images.
出处
《计算机应用研究》
CSCD
北大核心
2010年第2期784-787,共4页
Application Research of Computers
基金
广东省自然科学博士启动基金资助项目(7300090)
暨南大学引进人才科研启动基金资助项目(510061)
关键词
光学乐谱识别
谱线删除
乐符分割
文字提取
游程编码
optical music recognition
staff lines removal
music symbols segmentation
text extraction
run-length coding
