Music Score Binarization Based On Content Knowledge

In Optical Music Recognition (OMR), a kind of Optical Character Recognition that fo-cus on the recognition of music symbols, it is useful to build a system that allows the transformation of digitalized music scores into a machine-readable symbolic format like MusicXML. These systems start processing the documents by searching for music staves. Many works have been done in the area, although one of the most important pre-processing steps that precedes the staff finding step has always been overlooked: the binarization. This pre-processing consists in dividing the pixels of an image into two groups: the objects pixels (that get value 0) and the background pixels (that get value 1). Two major binarization techniques exist, the most common being the global thresholding in which a single intensity threshold is found for the whole image, dividing the pixels by intensity above or bellow that threshold. The adaptive methods, on the other hand, find a different threshold for each section of an image or even for each pixel. Although many papers have been published describing methods for binarization of images, there is currently no study or comparison between such methods applied to musi