Measurement of the accuracy of OMR based on the evaluation of errors by musicians

This thesis explores the landscape of Optical Music Recognition (OMR) with a focus on the development of an innovative accuracy measurement system. The work begins with an introduction to the research questions, objectives and methodology The literature review provides an overview of OMR and identifies existing tools and accuracy metrics.The methodology section outlines OMR systems, data collection and evaluation metrics. The prototype development is then detailed, covering ground truth data generation and difference detection. The novel accuracy measure is presented, including a questionnaire and error weighting, with the results defining the actual accuracy measure. The evaluation compares our method with edit distance accuracy and analyses the performance of the OMR system. The conclusion provides insights into research questions, acknowledges limitations and suggests future directions, presenting a comprehensive exploration of the challenges and opportunities of OMR.

Supervisors: Moreno Colombo, Edy Portmann

Student: Jérôme Vonlanthen

Project status: Finished

Year: 2024

Keywords: Optical Music Recognition, Windband, Machine learning

Document: Report