Multiple feature instance detection with density based clustering
Onalenna J. Makhura, John Woods · 2018
This paper describes an approach to local feature instance detection by treating this as a clustering problem. We use a density based clustering algorithm and a local feature detection algorithm. The paper also provides an approach to validating the results. Confidences are calculated for each cluster using core distances and intra-cluster distances to determine how valid the each cluster is for a particular value of the clustering algorithm parameter minPts. We used the statistical values of kurtosis and skewness to validate the choice of minPts and describe a way of using these values to make an informed choice of the proper value of minPts.