I've currently implemented mean shift clustering in python. I used matplotlib and I get the results I'm expecting on clustering on random input. However, I'm confused as to how one normally reduces the cluster for each point into a few clusters.
What I mean by this is mean shift outputs each point as a cluster, though many points may actually correspond to the same cluster. I had a few ways that I theorized this could be done, but It seemed odd to me that not a single tutorial I looked over actually explained how you would actually extract the unique clusters on output:
one could have a list of clusters for each point, and after shifting was done for each point, check if the newly created cluster is equal to the given point (or some close distance metric) add the original point of the shifted point to the cluster. Other wise create a new cluster as the pair [cluster_location: [cluster_points]]
one could have a hash (dictionary) where you hash the cluster positions and then if your current cluster position exists, add the un shifted point to that cluster, otherwise create a new cluster hash index.
I'm leaning toward doing B, but I would like to know what the standard method for cluster retrieval from Mean Shift actually is.