Limited Optical Music Recognition (for GSOC 2017)

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

Limited Optical Music Recognition (for GSOC 2017)

Jeevan
I wanted to build a limited OMR system for Musescore which can detect noteheads accurately. I believe this can speed up notation entry process into Musescore .
Features proposed
1.Accurate notehead detection (excluding the rythm)
2.Making a GUI which wherin a person will just have to select the particular rythm for each note in the sheet music.

I believe making such a system can help us gather the right amount of data in order make a totally automated OMR system.

please give your suggestions for the same .
Reply | Threaded
Open this post in threaded view
|

Re: Limited Optical Music Recognition (for GSOC 2017)

Liang Chen
Hi Jeevan, I've made some contributions last year to the lightweight OMR framework. I remember there is a place where we can directly plug in a notehead detector so that we can recognize note heads and barlines altogether. I tried some simple template matching approach before but it doesn't generalize to different scores. Probably you can try your approach using the same interface in the current framework and see the effect. I was busy with other things so didn't continue to contribute to this module. But I'm always interested in its progress and would like to keep in the discussion loop if need be.

Liang
Reply | Threaded
Open this post in threaded view
|

Re: Limited Optical Music Recognition (for GSOC 2017)

Jeevan
I don't think template matching would work very well as its not robust to rotation and other perspective transformations. What i had in mind was to Cascade a number of weak classifier together to get a strong classier (like the principle used in HAAR classfiers ) . Each of these classifiers should not miss out the note head but its okay if they detected some thing else in addition as we will only be taking the combined result of all the classifiers.