On the Potential of Fully Convolutional Neural Networks for Musical Symbol Detection

Abstract

Musical symbol detection on the page is an out- standing Optical Music Recognition (OMR) subproblem. We propose using a fully convolutional segmentation network to produce high-quality pixel-wise symbol probability masks.

Experiments on notehead detection show a very promising detection f-score of 0.98 with elementary detection methods.

Keywords

• potential
• fully
• convolutional
• neural
• networks
• musical
• symbol
• detection