Skip to Main content Skip to Navigation

Algorithmes de la morphologie mathématique pour les architectures orientées flux

Abstract : This thesis deals with the algorithms of Mathematical Morphology that can consider the pixels of an image as if they were a stream of data. We will show that a great number of algorithms of Mathematical Morphology can be described by data flows (streams) passing through the operating units. We will see that this approach can function on generic processors supporting a multi-media instruction set as well as on graphics cards. We propose to use the functional language Haskell for the description of algorithms operating on data flows. This allows us to describe the building blocks that are used to construct morphological algorithms. We apply these building blocks in the description of the most usually used algorithms (dilation/erosion, geodesic operations, distance function and levelings). This will also facilitate the porting of these algorithms onto several platforms. We propose an original mode of execution by macro blocks for the construction of morphological algorithms and we study in depth the transposition of this idea to SIMD architectures. We show that the use of macro blocks is interesting for multimedia architectures. We also show that the morphological algorithms proposed in this thesis reach better performances than standard implementations. Thus, a new field opens to algorithms we have developed here in real-time image processing applications. This thesis also explores the graphics processors and shows on experimental results that they are already powerful enough to compete with general processors.
Document type :
Complete list of metadatas

Cited literature [178 references]  Display  Hide  Download
Contributor : Ecole Mines Paristech <>
Submitted on : Thursday, August 5, 2010 - 5:29:33 PM
Last modification on : Friday, October 23, 2020 - 4:48:17 PM


  • HAL Id : pastel-00001879, version 1


Jaromír Brambor. Algorithmes de la morphologie mathématique pour les architectures orientées flux. Mathématiques [math]. École Nationale Supérieure des Mines de Paris, 2006. Français. ⟨NNT : 2006ENMP1368⟩. ⟨pastel-00001879⟩



Record views


Files downloads