A CNN-Based Optical Music Recognition Method
Jing Chen, Weixiang Gao, Kai Zhang · 2024
Optical Music Recognition (OMR) technology is an important means of music digitalization in the future. Today's society is developing in the direction of digitalization, virtualization, and intelligence, and the traditional music field is also deeply affected by it. In this paper, the OMR system mainly includes four parts: the preprocessing of music score images, the localization and deletion of spectral lines, the extraction and recognition of note motifs, and the reconstruction of music information. Finally, Pyqt5 is used to develop a complete system that is convenient for the input of score images, viewing descore images, text information, and playing audio to realize the conversion of scores. The extraction and recognition of note basis uses the direction gradient histogram and Convolutional Neural Networks (CNN) algorithm to train a model with high accuracy. The local connection, weight sharing, and pooling operation of CNN can effectively reduce the complexity of the network, reduce the number of training parameters, make the model have a certain degree of invariability to translation, distortion, and zoom, and have strong robustness and fault tolerance, and also be easy to train and optimize.