I'm looking for pointers to existing algorithms that take a number of position ellipses (centre, axis orientation, majr/minor axes) and cluster them together based on proximity to each other. The traditional way of clustering doesn't tend to use spatial size to inform the clusters.

An alternative way of thinking about it would be to assign a confidence value to the centre coordinates based on the size of the ellipse and use this to inform the cluster assignment.

If anyone can point me in the direction of doing this I would be grateful. I have searched and not found anything directly applicable.

  • $\begingroup$ I think expectation minimization should be possible to adapt to your case. When it determines the likelihood that a given point belongs to a set, it can consider the probability distribution of that data point as well, not only its mean $\endgroup$ – Aleksejs Fomins Aug 23 '18 at 10:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.