Cluster pixels in Binary Image from 2D-Histogram Data

What I am trying to achieve is to decompose a stl-model into sub-volumes based on the corresponding surface normals (or check if this is somehow possible or can be combined with other methods).

First I created a 2D-Histogram in Matlab based on the orientation of surface normals of a STL-File (Figure A, attached PNG).

For further processing I then transformed the Histogram data to a binary image. Now I am trying to cluster the binary data to achieve clusters like in figure C.

But best I can do for now is creating clusters based on the direct connectivity of the pixels (figure B) and I can't think of an approach to achieve this.

Maybe someone could give me a hint or algorithm to achieve the results like seen in figure C?

Thanks in advance!

Answer 1

[Ameer Hamza]

You can use regionprops(): https://www.mathworks.com/help/images/ref/regionprops.html to get each white region in your image. Then you will need to check the difference between x and y coordinates of region-centers to get the type of clustering you want.

[Timo Schmid]

I managed to cluster another histogram using DBScan-Function (delivered the best results) which you can see on the right (attached picture). For the right histogram it works fine.

But I can't manage to cluster along x/y direction in the left one because of the intersections... Any ideas on how to approach this?