Using Descriptor Counts in Clustering

Conclusion

• Descriptor counts contain useful probability information that can be leveraged for clustering large datasets.



• Performance gains versus N² appear feasible.



• More empirical results needed.
  (Got data?)