Many public-domain musical scores that are digitally available today have the form of images or scans. Approaches that turn such scans into symbolically encoded music representations are currently very sensitive to factors like printing variation, image quality and physical condition of the printed copy.
TROMPA will combine different Optical Music Recognition (OMR) systems, analogous to the method of (Hankinson, 2014), while also integrating global computer vision methods and strategic crowd input to increase transcription efficiency. At the same time, we shall work towards retrieval and analysis methods and a web-scale processing pipeline, based on promising prior work by PN and GOLD, that are largely robust to the effects of typical and inevitable OMR errors.
Hankinson, A.N. “Optical Music Recognition Infrastructure for Large-Scale Document Analysis”, PhD dissertation, McGill University, 2014.