OCR for sheet music

Question

Im considering doing a small project as a part of my masters for doing ocr just for sheetmusic instead of text.

I think PIL and Python would be fine for simple proof of concept O"notes"R.

My question is: Has anyone got any "dont do it with PIL use xyz instead" or something in that alley?

EDIT: My delicius links regarding the subject if anyone is interested: http://delicious.com/seet/DIKU-09b4%2Bb1

=========================================================================

EDIT2:

Actually, now I know a lot more about OCR for sheet music or OMR as it is called.

Within academia the area has been researched since late 60/early 70 and building an OMR system is not a simple task. To get a summary of the problems and the research until early 2000 you could read "The challenge of Optical Music Recognition" which is quite successful in drawing up the lines of the field.

Regarding existing software I know of at least these:

• Sharpeye
• Photoscore
• Smartscore X
• Capella Scan
• PdfToMusic

And my unscientific tests gave me the idea that photoscore was the most robust one.

For Opensource software Audiveris is the only complete thing I found and is written in Java.

Regarding my original question I am using Gamera. Gamera is an opensource tool for document image analysis which provides tools to do all the basic stuff needed for analysing images for recognition. Gamera has a python interface and the possibility to write c++ "toolkits". For example is it possible to download and use a staffline removal toolkit for gamera.

Answer 1

You may be interested in contributing to this project. Other than that, best of luck with your masters.

Answer 2

http://www.musitek.com/ has commercial offerings -- you can look at their user guide and other specifications to get some hints as to how to proceed.

Answer 3

My project ended with a report and some python software. Find it here:

http://preomr.appspot.com/

The gist of it is: It is hard to do good OMR and takes a lot of effort. I didn't have the time to do complete OMR (and it looks like it's not that needed after all).

I implemented a tool that can do preprocessing of sheet music before handing it to a OMR tool like Photoscore or the like. The preprocessing includes removal of lyrics and dynamics as this information is not needed for statistical analysis of the music in large music corpora