I would like to know if there is a best way to guess the number of cluster in online k-means. I'm trying to use kmeans to cluster lines in hough transform algorithm, but the number of cluster depends only of the image. I would like to parse the image only one time. So I guess that's is not possible, but I want to be sure.
Thank you !!!
One simple way would be a threshold.
Namely if the new element is farther than any cluster above certain threshold you'd open a new cluster with it as a centroid.