Skip to Main content Skip to Navigation

Régularisation d'Images Multivaluées par EDP et Applications

David Tschumperlé 1
1 ACACIA - Knowledge acquisition for aided design through agent interaction
CRISAM - Inria Sophia Antipolis - Méditerranée
Abstract : We are interested in PDE-based approaches for vector-valued image regularization, and its applications for a wide class of interesting image processing problems. The comparative study of existing methods allows us to propose a common mathematical framework, better adapted to understand the underlying diffusion geometry of the regularization processes, as well as design corresponding numerical schemes. Thus we develop a new multivalued image regularization approach that verifies important geometric properties. It can be used in a large range of regularization-related applications. We also tackle the problem of constrained regularization and propose a specific variational
formalism unifying in a common framework, the equations acting on direction features~: unit vectors, rotation matrices, diffusion tensors, etc.
Proposed solutions are analyzed and used with success to solve applications of interest, such as color image regularization and interpolation, flow visualization, regularization of rigid motions
estimated from video sequences, and aided reconstruction of coherent fibers network models in the white matter of the brain, using DT-MRI imaging.
Complete list of metadatas

Cited literature [199 references]  Display  Hide  Download
Contributor : David Tschumperlé <>
Submitted on : Wednesday, February 12, 2003 - 2:59:57 PM
Last modification on : Thursday, March 5, 2020 - 4:50:02 PM
Long-term archiving on: : Monday, September 20, 2010 - 11:22:54 AM


  • HAL Id : tel-00002396, version 2



David Tschumperlé. Régularisation d'Images Multivaluées par EDP et Applications. Interface homme-machine [cs.HC]. Université Nice Sophia Antipolis, 2002. Français. ⟨tel-00002396v2⟩



Record views


Files downloads