Category CS P08 Optical Music Recognition of Printed Music Score

Abstract Computer vision is one of the most interesting areas in Computer

Science. For many musicologists, manually encoding musical information is

expensive and time consuming. Since the 1950s, researchers have

attempted to turn sheet music into digitized files through the use of a

computer. There has been progress and success in Optical Character

Recognition (OCR) to the point it can be used commercially. However,

Optical Music Recognition (OMR) has been studied since late 1959 but its

success has been minor compared to that of OCR. Difficulty of OMR has

deterred researchers from further research. Nonetheless, OMR can be

just as efficient as OCR and is a promising tool for musicians and

musicologists. Some of the benefits of OMR are listed.



1. Read in a score of music and then play it out, using an appropriate

audio-music-producing program.

2. Read existing score to transfer the data to storage or transmission

media.

3. Read in a newly engraved piece of music and proofread it for

syntactic and other errors.



For the project, Mathematica 8 was used to implement and test

algorithms. Eight pages of Mozart’s “String Quintet in B-flat Major, K. 174”

score were scanned in full 300 dots per inch (dpi) at full page mode and

with 1 bit per pixel. To optimize performance, they were saved as TIFF

(Tagged Image File Format) in a 2550 by 3300 pixel format. After importing

them to Mathematica, each page was cut into individual measures using

segmentation techniques. Using pattern recognition techniques, various

types of annotations like notes and rests were assigned sounds. ASCII file

was used as an intermediate to store information like pitches, duration,

and note names. Then, the file was converted to MIDI file format, which

could be played on Windows Media Player.



This experiment was attempted to read printed music scores using

existing pattern recognition algorithms on Mathematica 8. The emphasis

was on using more efficient methods such as projection matching and size

classification, rather than brute force ones such as pattern matching.

Although many improvements are still needed to make a general OMR

system, the benefit of OMR is enormous. With electronic synthesizer

sound coming closer to real instruments, OMR can bring full orchestra

sound to a composer from a conceptual stage and also make it easier for

composers to have their music printed. Musicologists can concentrate on

their research rather than wasting time inputting the data into the

computer.



Bibliography 1) http://www.npcimaging.com/thesis/Chap1.html 2)

http://journal.code4lib.org/articles/84 3) http://www.jstor.org/pss/832471
First Previous Next Last