Cov-HGMEM: An Improved Hierarchical Clustering Algorithm

摘要

In this paper we present an improved method for hierarchical clustering of Gaussian mixture components derived from Hierarchical Gaussian Mixture Expectation Maximization (HGMEM) algorithm. As HGMEM performs, it is efficient in reducing a large mixture of Gaussians into a smaller mixture while still preserving the component structure of the original mode. Compared with HGMEM algorithm, it takes covariance into account in Expectation-Step without affecting the Maximization-Step, avoiding excessive expansion of some components, and we simply call it Cov-HGMEM. Image retrieval experiments indicate that our proposed algorithm outperforms previously suggested method.