Skip to content

Latest commit

 

History

History
43 lines (27 loc) · 1.09 KB

README.md

File metadata and controls

43 lines (27 loc) · 1.09 KB

Notes Recognition

An image processing software that gets as input an image of some notes, and as an output plays the melody of those notes

Theory + Algorithms

  • Hough Transform
  • Simple NN for recognizing the bars

Prequesites

  • OpenCV

Building & Running the Code

cd /.../Notes-Recog-master
make
./notes

What we did:

We seperated our project into 2 main parts:

Step 1) - Identifying the individual bars

We ran first a binarization on the image.

Next we ran a hough transform for straight lines, using an algorithm we have seen in class.

We eliminated all diagonal lines by calculating the slope and increasing the threshold (most music sheets contain perpendicular lines only)

We then hunted for the rows that we wanted (a collection of bars). This was done by finding the large gaps between the horizontal lines, and then taking the inverse - we wanted rows not gaps.

Finally, we sliced each row into bars using the vertical lines found. each bar image was cropped and numbered, and was passed onto stage 2.

Step 2) - Converting the bars' notes into a format we can play