Musical Note Position and Duration Recognition Model in Optical Music Recognition Using Convolutional Neural Network

Abstract

This study aims to solve Optical Music Recognition (OMR) problems using a non-End-to-End (non-E2E) approach. Therefore, separate models for Position Recognition (PR) and Duration Recognition (DR) are constructed, with both employing Convolutional Neural Networks (CNN). In terms of constructing a non-E2E architecture to solve OMR problems, this study obtains superior evaluation results compared to previous research, with the PR and DR models achieving accuracies of 97.88% and 99.23%, respectively. In addition, this study employs template matching in conjunction with several supplementary tasks to identify the positions of musical notes and generate the corresponding note sequences in the intended reading format. Our Optical Music Recognition (OMR) system can accomplish comparable results to the E2E architecture by utilizing these techniques.