PUMA
Istituto di Scienza e Tecnologie dell'Informazione     
Bedini L., Tonazzini A., Minutoli S. A neural architecture for simultaneous MAP image restoration and ML estimation of edge-preserving Gibbs priors. Technical report, 1996.
 
 
Abstract
(English)
This paper proposes an iterative procedure that simultaneously computes the MAP restoration of an image and the ML estimation of the unknown model parameters, when piecewise smooth MRF models with an explicit, constrained line process are considered. The method consists of a mixed-annealing algorithm for the maximization of the posterior distribution with respect to the image field, periodically interrupted to compute a new set of parameters. Owing to the presence of an explicit line process, we define these new parameters as those that maximize the conditional prior distribution of the lines given the intensities, evaluated on the current estimate of the whole image field. The computation of the expectations involved in the ML estimation can thus be performed by summation over the binary line elements alone, with a strong reduction in computational complexity. Moreover, when a model with non-interacting discontinuities (e.g. the weak membrane) is considered, the expectations can be computed analytically, with a further reduction in the computational load. Finally, a neural architecture, based on a Hopfield net interconnected with a Boltzmann Machine, which also holds for the general case of self-interacting discontinuities, is proposed for a possible analogue implementation of the whole procedure.
Subject Neural architecture
I.4.4 Restoration
I.4.6 Segmentation: Edge andfeature detection
I.2.6 Learning: Connectionis and neural nets
I.2.6 Learning: Parameter learning
G.3 PROBABILITY AND STATISTICS: Probabilistic algorithms (including Monte Carlo)


Icona documento 1) Download Document PS


Icona documento Open access Icona documento Restricted Icona documento Private

 


Per ulteriori informazioni, contattare: Librarian http://puma.isti.cnr.it

Valid HTML 4.0 Transitional