Optical music recognition is closely related to optical character recognition, but instead of characters it recognizes notes on a note sheet. The program handles scanned in images as well as photographed images at different angles and with some distortion. Geometric correction of distorted note lines was made to straighten up the note lines, and then a horizontal projection was made to estimate the note lines positions for calculation of the pitch. To classify the notes a lot of morphologic operations, such as dilate and erode, were made as well as structure classifications of the horizontal and vertical bars. The node heads, horizontal and vertical bars were added together in one images and then grouped if they intersected each other, and therefore probably would be a note of some kind. These notes were then sorted and classified and then written out as accords.
Advanced methods for image processing and analysis in the area of the project task.
Full course information: