Unsupervised hierarchical clustering based on sequential partitioning and merging
Sheng‐Jyh Wang · 2016
In this talk, we present an unsupervised hierarchical clustering method based on the split-and-merge scheme. In the splitting phase, we sequentially partition the feature space of the given data into smaller cells so that the probability distribution of the feature points within each cell follows a Gaussian function. In the merging phase, we sequentially merge cells of similar feature property into larger cells to construct a hierarchical description of the data. This hierarchical representation is very efficient and effective in describing the inherent structure of the data, especially data of high dimension. An application of this unsupervised hierarchical clustering algorithm in image segmentation is also presented to demonstrate the feasibility of this new approach.