Joint Model-Pixel Segmentation with Pose-invariant Deformable Graph-Priors

Abstract : This paper proposes a novel framework for image segmentation through a unified model-based and pixel-driven integrated graphical model. Prior knowledge is expressed through the deformation of a discrete model that consists of decomposing the shape of interest into a set of higher order cliques (triplets). Such decomposition allows the introduction of region-driven image statistics as well as pose-invariant (i.e. translation, rotation and scale) constraints whose accumulation introduces global deformation constraints on the model. Regional triangles are associated with pixels labeling which aims to create consistency between the model and the image space. The proposed formulation is pose-invariant, can integrate regional statistics in a natural and efficient manner while being able to produce solutions unobserved during training. The challenging problem of tagged cardiac MR image segmentation is used to demonstrate the performance potentials of the method.
Liste complète des métadonnées

Cited literature [14 references]  Display  Hide  Download

https://hal.inria.fr/hal-00856955
Contributor : Bo Xiang <>
Submitted on : Monday, September 2, 2013 - 4:54:02 PM
Last modification on : Tuesday, February 26, 2019 - 10:12:33 AM
Document(s) archivé(s) le : Tuesday, December 3, 2013 - 6:00:16 AM

File

paper924.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-00856955, version 1

Collections

Citation

Bo Xiang, Jean-François Deux, Alain Rahmouni, Nikolaos Paragios. Joint Model-Pixel Segmentation with Pose-invariant Deformable Graph-Priors. 16th International Conference on Medical Image Computing and Computer Assisted Intervention, Sep 2013, Nagoya, Japan. ⟨hal-00856955⟩

Share

Metrics

Record views

445

Files downloads

197