Optical music recognition basics

Question

0 votes

Hello everyone, For a few months i have been making researches on Optical music recognition which somehow may be close to OCR.

An Optical music recognition software basically reads the computer written piano sheets like this one and translate it into notations i can then match to play the piano peace through recorded sounds.

Here are few things and concepts i have reached till now :

Steps of implementation

   * Staff line detection : which is removing the 5 lines in the Sheets.
   * Removing any noise
   * Detecting all the symbols noted.
   * Somehow matching symbols to staff lines and knowing the note pitch .

2. Algorithms for reaching these steps

   * Staff line detection : 
       1- Morphological operations like in this link (knowing it is not Mat lab.)
       <http://www.answers.opencv.org/question/53293/how-to-remove-line-on-music-sheet/>
       2- flood fill algorithm 
       3- Hough transform
       4- de facto

Here comes my question :

In step 2 i have to detect symbols, this process may be called as layout analysis or zoning. then passing each symbol to a classifier to identify each symbol ... this may be done according to my little knowledge using fast Fourier transform for matching pictures.

Anyhow what algorithms can i use for implementing such classifier?
What algorithms may help me scan symbol by symbol .. or area by area like a mask scanner or something like that before providing classifier or matching tool like FFT or any other concept ?
How can i introduce concepts like neural networks, support vector machines and machine learning in such project knowing it is not a hand written notes it a simple computerized or printed.
Please Guide me ... if you need more clarification please ask i will provide you with more details about my needs and provide links, books or articles that may help.

Thanks.