Skip to Main content Skip to Navigation

Architectures parallèles pour la morphologie mathématique géodésique

Abstract : This thesis adresses the study of dedicated hardware for computer-aided vision. We focus on mathematical morphology operators for they are well suited for industrial purpose: generic, robust, easy to implement. These operators are based on data wavefront propagation according to geodesy and idempotence: watershed, reconstruction, labeling, etc. Their efficiency is shown on a real application cases and leads to a list of operators we will implement. Two original approaches are investigated. First, a massively parallel cellular automaton is described. With such an architecture, every data wavefrontcan be processed simultaneously. Associative mecanisms are used in order to reduce the synchronization contraints between the processors. Then, a pipeline architecture is depicted. It is more adapted to an industrial project context. This architecture implements generic algorithms based on a data driven image scan. We emphasis the way data control and flow parallelism are exploited since they are the key difficulties to overcome in non uniform processing. This architecture, named SPIDDO, requires about 40 ms to perform the watershed transform when clock frequency is set to 25 MHz.
Complete list of metadata

Cited literature [165 references]  Display  Hide  Download
Contributor : Lucie Torella <>
Submitted on : Monday, June 23, 2003 - 10:08:22 AM
Last modification on : Friday, December 11, 2020 - 8:28:03 AM
Long-term archiving on: : Friday, April 2, 2010 - 8:11:11 PM


  • HAL Id : tel-00003040, version 1




D. Noguet. Architectures parallèles pour la morphologie mathématique géodésique. Micro et nanotechnologies/Microélectronique. Institut National Polytechnique de Grenoble - INPG, 2002. Français. ⟨tel-00003040⟩



Record views


Files downloads