Cluster-Based Implementation of a Morphological Watershed Algorithm for Parallel Classification of Multichannel Images

Due to the large data volumes often associated with multichannel imagery in many application domains, including satellite imaging and aerial reconnaissance, there is a clear need for parallel algorithms able to process these high-dimensional data sets quickly enough for practical use. This paper describes a parallel implementation of a multichannel classification algorithm that naturally combines the information provided by all dimensions. The parallel algorithm first defines multichannel morphological operations, and then uses these operations to extend the morphological watershed transformation to multidimensional images. The proposed technique has been implemented on Thunder-head, a Beowulf cluster at NASA´s Goddard Space Flight Center, and tested using remotely sensed data collected by the Jet Propulsion Laboratory´s Airborne Visible Infra-Red Imaging Spectrometer (AVIRIS).