Contributions à la description de signaux, d'images et de volumes par l'approche probabiliste et statistique

Olivier Alata 1
XLIM - XLIM, Université de Poitiers
Abstract : The main elements described in this document are: - The relevance of $\phi_\beta$ information criterion which allows the learning of $\beta$ parameter whatever is the model selection problem solved with an information criterion, possibility which has been illustrated in different applicative contexts (linear prediction support and dimension of the model used for $\dot VO_2$ cinetics). - An estimation method for histograms in order to approach in a non-parametric manner the distribution of samples and its use for law recognition in a supervised way. - A method called ``comparative descendante'' which allows to find the best combination of parameters without testing all combinations in order to describe datas. This method has been illustrated on the problem of selection of 1-d and 2-d linear prediction supports. - The implementation of strategies for chosing some models in various contexts like for TEP volumes and Poisson and Gauss mixture laws, or the color spaces and the multidimensional gaussian mixture models. - The study of multicanal complex prediction linear models on images represented in color spaces which separate intensity from chromatic part and their use for texture caracterization, texture classification and color texture image segmentation. - Some contributions for image segmentation: the optimization of an unsupervised segmentation method for textured images in gray levels ; a supervised segmentation method for textured images represented in psychovisual color spaces based on multicanal complex linear prediction error ; tacking into account geometrical and topological information from the region field in Gibbs distribution in order to propose an ``high-level'' 3-d segmentation method based on point process. - The use of MCMC methods for parameter estimation, 2-d and 3-d segmentation methods or the simulation of processes. And many other things that you will find during the reading of this document ...
Document type :
Habilitation à diriger des recherches
Signal and Image processing. Université de Poitiers, 2010
Contributor : Olivier Alata <>
Submitted on : Thursday, March 3, 2011 - 11:04:31 AM
Last modification on : Monday, October 13, 2014 - 3:43:25 PM


  • HAL Id : tel-00573224, version 1



Olivier Alata. Contributions à la description de signaux, d'images et de volumes par l'approche probabiliste et statistique. Signal and Image processing. Université de Poitiers, 2010. <tel-00573224>




Consultation de
la notice


Téléchargement du document