HOGMep: Variational Bayes and higher-order graphical models applied to joint image recovery and segmentation

Aurélie Pirayre (1) , Yuling Y. Zheng (2) , Laurent Duval (1) , Jean-Christophe Pesquet (3, 4, 5)

1 IFPEN - IFP Energies nouvelles
2 CRL - IBM-China Research Laboratory
3 LIGM - Laboratoire d'Informatique Gaspard-Monge
4 CVN - Centre de vision numérique
5 GALEN - Organ Modeling through Extraction, Representation and Understanding of Medical Image Content

Variational Bayesian approaches have been successfully applied to image segmentation. They usually rely on a Potts model for the hidden label variables and a Gaussian assumption on pixel intensities within a given class. Such models may however be limited, especially in the case of multicomponent images. We overcome this limitation with HOGMep, a Bayesian formulation based on a higher-order graphical model (HOGM) on labels and a Multivari-ate Exponential Power (MEP) prior for intensities in a class. Then, we develop an efficient statistical estimation method to solve the associated problem. Its flexibility accommodates to a broad range of applications, demonstrated on multicomponent image segmentation and restoration.

Mots clés: variational Bayes; higher-order graphical models; multicomponent images; segmentation; restoration

Domaines: Traitement du signal et de l'image [eess.SP]; Bio-informatique [q-bio.QM]; Optimisation et contrôle [math.OC]; Analyse de données, Statistiques et Probabilités [physics.data-an]

Aurélie Pirayre, Yuling Y. Zheng, Laurent Duval, Jean-Christophe Pesquet. HOGMep: Variational Bayes and higher-order graphical models applied to joint image recovery and segmentation. 2017 IEEE International Conference on Image Processing (ICIP), Sep 2017, Beijing, China. ⟨10.1109/ICIP.2017.8296988⟩. ⟨hal-01862840⟩