Parallel hierarchical clustering algorithms on processor arrays with a reconfigurable bus system

摘要

Clustering techniques are usually used in pattern recognition, image segmentation and object detection. Let N be the number of patterns and M be the number of features of each pattern and N _̆ M. In this paper, we first design two O(1) time basic operations for concentrating all nonempty data of size N and computing the proximity matrix using N × N and N × N × M processors, respectively. Then, based on these two operations, a constant time parallel hierarchical clustering algorithm is proposed on a 3-D processor array with reconfigurable bus system using N4 processors. Then, by reducing the number of processors by a factor of N, an O(log2 N) time algorithm for this problem is also derived. Note that no one had ever obtained a constant time algorithm for this problem on the existing parallel computation models.