Exploiting the Two-Dimensional Nature of Agnostic Music Notation for Neural Optical Music Recognition

State-of-the-art Optical Music Recognition (OMR) techniques follow an end-to-end or holistic approach, i.e., a sole stage for completely processing single-staff section image and retrieving the symbols that appear therein. Such recognition systems are characterized by not requiring exact alignment between each staff their corresponding labels, hence facilitating creation retrieval of labeled corpora. Most commonly, these approaches consider agnostic music representation, which characterizes shape height (vertical position in staff). However, this double nature is ignored since, learning process, two features treated as single symbol. This work aims to exploit trademark differentiates notation from other similar domains, such text, introducing novel approach solve OMR task at staff-line level. We Convolutional Recurrent Neural Network (CRNN) schemes trained simultaneously extract information propose different policies eventually merging them actual neural The results obtained corpora monophonic early manuscripts prove our proposal significantly decreases error figures ranging 14.4% 25.6% best-case scenarios when compared baseline considered.