A recursive PVM implementation of an image segmentation algorithm with performance results comparing the HIVE and the Cray T3E

Abstract

A recursive PVM (Parallel Virtual Machine) implementation of a high quality but computationally intensive image segmentation approach is described and the performance of the algorithm on the HIVE and on the Cray T3E is contrasted. The image segmentation algorithm, which is designed for the analysis of multispectral or hyperspectral remotely sensed imagery data, is a hybrid of region growing and spectral clustering that produces a hierarchical set of image segmentations based on detected natural convergence points. The HIVE is a Beowulf-class parallel computer consisting of 66 Pentium Pro PCs (64 slaves and 2 controllers) with 2 processors per PC (for 128 total slave processors) which was developed and assembled by the Applied Information Sciences Branch at NASA's Goddard Space Flight Center. The Cray T3E is a supercomputer with 512 available processors, which is installed at the NASA Center for Computational Science at NASA's Goddard Space Flight Center. Timing results on Landsat Multispectral Scanner data show that the algorithm runs approximately 1.5 times faster on the HIVE, even though the HIVE is some 86 times less costly than the Cray T3E.