CODEWITHSUNDEEP
image-processing
Keegan Jay
Keegan Jay

Reputation: 960

Detecting location of music note dots in a sheet music image

I want to start a project that uses a very basic form of optical music recognition.

For those who understand sheet music: Unlike other OMR projects, the only information which needs to be extracted is the order and pitch values of each note in a bar. Quarter notes, half notes and whole notes need to be distinguished. Shorter notes can acceptably be interpreted as quarter notes. Dots on notes can be ignored. Dynamics markings are not important

For everyone: Strictly speaking I need to find the locations of each of the following...

enter image description here

... in a sample image like this... enter image description here

I have no experience in image processing so a basic, conceptual explanation of what technique or set of techniques are used to achieve this would be greatly appreciated.

Upvotes: 4

Views: 2630

Answers (2)

roschach
roschach

Reputation: 9366

I did something similar to your work and trust me it is a complete mess.

Howeverr, for the pitch for each note you extract the head from the rest and calculate the baricentre and compare its position to the position of lines calculated with Hough transform as already said (assuming that the lines are already straight: if not i think you can use the Fourier Transform).

For the duration you need a classification algorithm.

Upvotes: 0

Andrey Rubshtein
Andrey Rubshtein

Reputation: 20915

I would do the following:

Extract the line locations using Hough transform. (You get the angle as well). Crop each group of lines (5 lines), and process individually.

For each group of lines, you know the angle of the lines, so you can get the locations of the vertical small lines that separate the bars. Search again in Hough space, but with specific angle. (The original + 90). Crop each bar and process individually.

For each bar, use template matching on the possible notes (Quarter, Half,etc..)

Upvotes: 2

Related Questions