An Improved Method for Estimating the Modes of the Probability Density Function and the Number of Classes for PDF-based Clustering
This paper proposes an improvement of the estimation of the modes of the Probability Density Function (PDF) in clustering procedures. The k nearest neighbours are excluded when the PDF is estimated trying to avoid parasitic modes of the PDF estimation. The number of detected modes is analyzed using the bootstrap technique. The number of clusters is equal to the most frequent number of modes obtained when resampling the data (if the frequency is greater than 50%). The method to estimate the number of clusters is illustrated by an example in image processing.
KeywordsProbability Density Func Probability Density Function Gaussian Mixture Model Smoothing Parameter Cluster Procedure
Unable to display preview. Download preview PDF.