Parallel architectures for wavelet transforms
Mounir Hamdi · 2002
Discrete wavelet transforms have been shown to be a viable tool for multiscale image processing. However, their implementations on a sequential machine hinders their potential use for many applications because of their computational intensive nature. A parallelization of a 2D discrete wavelet transforms has been implemented on two novel parallel architectures: an array of processors with pipelined optical buses, and a 3D mesh architecture. The implementation of the DWTs on these architectures are analyzed. Then, their performance are shown to be optimal or near-optimal, and compare favorably with other related parallel architectures. The important consideration of implementing a DWT are shown to be the convolution and the multiscale processing. An array of processors with pipelined optical buses a 3-D mesh are shown to be quite suitable for these parallel data movement operations, and hence, seem to be promising architectures for DWTs applications.