激光与光电子学进展, 2019, 56 (6): 061006, 网络出版: 2019-07-30
基于多维局部二值模式和XGBoost的轻量谱线删除法 下载: 1157次
Lightweight Staff Removal Method Based on Multidimensional Local Binary Pattern and XGBoost
数字图像处理 局部二值模式 光学乐谱识别 谱线删除 digital image processing local binary pattern optical music recognition spectrum line deletion
摘要
由于在手写乐谱中搜索谱线位置比较困难,为提高乐谱谱线删除算法的稳健性,提出了一种基于多维局部二值模式识别和XGBoost模型的手写乐谱谱线删除方法。根据乐谱图像的特点,设计并改进局部二值模式算子,提取乐谱图像中的多维局部二值模式特征算子,组成高维特征向量,再选择最优的XGBoost模型来识别乐谱谱线位置,进而删除谱线。研究结果表明,该方法在测试数据上的F -measure为97.19%,说明其具有很高的准确率和召回率;而在三个不同子测试集上的F -measure分别为96.43%,98.36%和96.79%,说明其具有很好的稳健性。相比已有的轻量谱线删除算法,该方法的F -measure有所提升。
Abstract
It is difficult to search the spectrum line in handwritten music spectrum, so in order to improve the robustness of the handwritten music spectral line deletion algorithm, a method based on multidimensional local binary pattern recognition and XGBoost model is proposed. The local binary pattern operator is designed and improved based on the characteristics of music score image, and from which the multidimension local binary pattern feature operator is extracted. Therefore, a high-dimensional feature vector is formed and the optimal XGBoost model is selected to identify the music spectral line location, then the line is deleted. The research results show that F-measure of this method is 97.19% on the test data, which illustrates that the method has a high accuracy and recall rate. F-measure is 96.43%, 98.36% and 96.79% respectively on three different test subsets, which shows that it has good robustness. Compared with existing lightweight spectrum line deletion algorithm, the F-measure of this method is relatively improved.
吴天龙, 李锵, 关欣. 基于多维局部二值模式和XGBoost的轻量谱线删除法[J]. 激光与光电子学进展, 2019, 56(6): 061006. Tianlong Wu, Qiang Li, Xin Guan. Lightweight Staff Removal Method Based on Multidimensional Local Binary Pattern and XGBoost[J]. Laser & Optoelectronics Progress, 2019, 56(6): 061006.