Parallel architectures for wavelet transforms

Mounir Hamdi*

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

6 Citations (Scopus)

Abstract

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.

Original languageEnglish
Title of host publicationProceedings - 1993 Computer Architectures for Machine Perception, CAMP 1993
EditorsLarry S. Davis, Magdy A. Bayoumi, Kimon P. Valavanis
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages376-384
Number of pages9
ISBN (Electronic)0818654201, 9780818654206
DOIs
Publication statusPublished - 1993
Externally publishedYes
Event1993 Computer Architectures for Machine Perception, CAMP 1993 - New Orleans, United States
Duration: 15 Dec 199317 Dec 1993

Publication series

NameProceedings - 1993 Computer Architectures for Machine Perception, CAMP 1993

Conference

Conference1993 Computer Architectures for Machine Perception, CAMP 1993
Country/TerritoryUnited States
CityNew Orleans
Period15/12/9317/12/93

Keywords

  • Algorithms
  • Convolution
  • Parallel Processing
  • Wavelet Transforms
  • pipelining

Fingerprint

Dive into the research topics of 'Parallel architectures for wavelet transforms'. Together they form a unique fingerprint.

Cite this