A Novel Clustering Algorithm Based on Bayesian Sequential Partition and Its Application in Image Segmentation

In this work, we propose a novel clustering algorithm based on Bayesian Sequential Partition (BSP) and the spectral clustering algorithm. Since the BSP is capable of providing much more accurate density estimates when the sample space is of moderate to high dimension, the proposed clustering algorithm is believed to be superior to available ones when dealing with high dimensional problems. To demonstrate the effectiveness and efficiency of our novel clustering method, we have compared our new algorithm with the original spectral clustering algorithm and the meanshirt method in several image segmentation problems. Based on the results from our experimental studies, it can be shown that our new algorithm is indeed much more powerful.