Istituto di Scienza e Tecnologie dell'Informazione     
Tonazzini A., Bedini L. Degradation identification and model parameter estimation in discontinuity-adaptive visual reconstruction. Published in: Advances in Imaging and Electron Physics , P.W. Hawkes ed., Academic Press, 2002, Vol. 120, pp.193-284. www.academicpress.com/aiep/, Technical report, 2001.
This paper describes our recent experiences and progress towards an efficient solution of the highly ill-posed and computationally demanding problem of blind and unsupervised visual reconstruction. Our case study is image restoration, i.e. deblurring and denoising. The methodology employed makes reference to edge-preserving regularization. This is formulated both in a fully Bayesian framework, using a MRF image model with explicit, and possibly geometrically constrained, line processes, and in a deterministic framework, where the line process is addressed in an implicit manner, by using a particular MRF model which allows for self-interactions of the line and an adaptive variation of the model parameters. These MRF models have been proven to be efficient in modeling the local regularity properties of most real scenes, as well as the local regularity of object boundaries and intensity discontinuities. In both cases, our approach to this problem attempts to effectively exploit the correlation between intensities and lines, and is based on the assumption that the line process alone, when correctly recovered and located, can retain a good deal of information about both the hyperparameters that best model the whole image and the degradation features. We show that these approaches offer a way to improve both the quality of the reconstructed image, and also the estimates of the degradation and model parameters, and significantly reduce the computational burden of the estimation processes.
Subject I.4.4 [Image Processing]: Restoration
I.4.6 [Image Processing]: Segmentation . Edge and feature detection
G.1.3 [Numerical Linear Algebra]: Sparse, structured and very largesystems (direct and iterative methods)
G.3 [Probability and Statistics] . Probabilistic algorithms (includingMonte Carlo)
I.2.6 [Artificial Intelligence]: Learning: Connectionism and neural nets
Parameter learning

Icona documento 1) Download Document PDF

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