Multi-Resolution Wavelet Decomposition on the MasPar Massively Parallel System

Wavelet transforms have started to receive increasing attention due to their wide applications in image and signal processing, and due to the recent discovery of a fast sequential algorithm for discrete wavelet transforms by Mallat[Mal89]. In this paper we present a mapping of the wavelet operations onto the MasPar massively parallel single instruction multiple data (SIMD) computer [Bla90], [Mas92], [Nic90] and show that our MasPar algorithm is both (asymptotically) processor optimal and space optimal. We also present and discuss the implications of our timing measurements collected on a MasPar MP-1 and a MasPar MP-2, each with 16,384 processors, for variations of our algorithm. These algorithmic variations are tested using remotely sensed data, and their respective timings are compared and analyzed.