Skip to Main content Skip to Navigation

Parallélisation d'un algorithme d'appariement d'images quasi-dense

Abstract : In every field of scientific and industrial research, the extension of the use of Computer Science has resulted in an increasing need for computing power. Within this context, parallel computing appears as an important tool in order to allow the exploitation of many microprocessors working together. The field to which parallel computing will be applied in this thesis is computer graphics, more preciselly in synthesis of virtual scenes from real images. In this process, the pixel matching step is the most computing expensive task. It deals with creating couples of pixels which represent the same point over two source scenes surface. The main proposal of this thesis is a new version of the "quasi-dense" pixel matching algorithm (also called propagation algorithm). This new algorithm is a parallel version that relies on a global adaptative strategy. This version must be able to preserve the final result quality obtained by the sequential version. Furthermore, it must also reduce its execution time in a cluster-oriented programming context.
Complete list of metadata
Contributor : Thèses Imag <>
Submitted on : Friday, January 30, 2004 - 5:33:43 PM
Last modification on : Friday, November 6, 2020 - 4:39:27 AM
Long-term archiving on: : Friday, April 2, 2010 - 7:11:29 PM


  • HAL Id : tel-00004434, version 1




Luiz Gustavo Leão Fernandes. Parallélisation d'un algorithme d'appariement d'images quasi-dense. Interface homme-machine [cs.HC]. Institut National Polytechnique de Grenoble - INPG, 2002. Français. ⟨tel-00004434⟩



Record views


Files downloads