I am trying to find ways to segment a image randomly, but drawn from a probalistic distribution of pre-determined areas to be cut through. First thought was to pick random points and run Covex hull algorithms to generate the polygons. However that still leaves the problem of probabilistically making sure the polygons pass through/segment/cut certain areas as an optimization problem.
I was reminded of uniform random distribution Birdson's algorithm seemed to have a good idea. Essentially, I'm thinking I can borrow the idea of building from existing samples to form the random points, however, trying to figure out a way to use the preset salient segments for the distribution is stumping me. (distance from the salient parts correlate with probability?) Any ideas/suggestions?