Chair: Alfons H. Salden, University Hospital Utrecht, The Netherlands
Alfons H. Salden, INRIA Sophia Antipolis (France)
Image formation is quantised by imposing an image induced connection and computing the associated torsion and curvature in terms of differential or integral invariants. Imposing invariance of the image formation under the classical group of movements slotmachines reading out the torsion and the curvature are shown to locate endpoints and other type of interesting topological objects. Requiring instead invariance under the group of anamorphoses and the group of diffeomorphisms of the image only ridges and ruts can be identified through a non-local topological or integral geometric operation.
Bogdan Cramariuc, Tampere University of Technology (Finland)
Ioan Tabus, Tampere University of Technology (Finland)
Moncef Gabbouj, Tampere University of Technology (Finland)
The purpose of the present paper is to analyze how the predictive distribution estimated using a set of context dependent nonlinear adaptive predictors can be used to localize edges in graylevel images. Since the adaptive predictors have the potential of learning repetitive structure, as those characteristic to certain textures, our predictive edge detection scheme may be a practical way to conceal the relative high contrast of certain texture regions.
C. Ju, Memorial University of Newfoundland (Canada)
C. Moloney, Memorial University of Newfoundland (Canada)
One problem in processing Synthetic Aperture Radar (SAR) images is the presence of speckle noise which is multiplicative in the sense that the noise level increases with the magnitude of radar backscattering. In this paper, an edge-enhanced filtering method is presented which is based on a ratio-based edge detector used in conjunction with an iterative application of a modified Lee filter. Test results obtained by applying this method to synthetic images corrupted by SAR speckle show that the edge-enhanced modified Lee filter has the ability to remove speckle in both low and high variance regions while retaining the sharpness of edges even after several iterations. As such, this filtering method may be useful in SAR image segmentation and classification applications.
Mark Mertens, Vrije Universiteit Brussel (Belgium)
Hichem Sahli, Vrije Universiteit Brussel (Belgium)
Jan Cornelis, Vrije Universiteit Brussel (Belgium)
In this paper we describe a nonlinear criterion designed for the detection of changes ("edges") in signal or image properties in a framework that we call the distinction evidence method. It was introduced as a generic feature extraction tool for image modeling. We show its capabilities when applying it to segmentation, texture border finding and object correlation problems.
Marco Storace, University of Genoa (Italy)
Mauro Parodi, University of Genoa (Italy)
Carlo S. Regazzoni, University of Genoa (Italy)
A methodological approach to the definition of nonlinear circuits for real-time image processing is presented. The image processing problem is formulated in terms of the minimization of a functional based on the Markov Random Fields (MRFs) theory. The terms of such a functional are related to the co-contents of proper nonlinear multiterminal resistors, thus reporting the minimization process to the achievement of an equilibrium solution in a circuit made up of these multiterminal resistors and of linear capacitors. Coupled image restoration and edge extraction in the presence of additive gaussian noise are the specific problems addressed in this paper.
Vadim Mottl, Tula State University (Russia)
Alex Kostin, Tula State University (Russia)
Ilya Muchnik, Rutgers University (U.S.A.)
Under the generalized smoothing of a signal, a wider class of operations is understood than the plain suppression of noise. We apply this term to all the processing problems which can be interpreted as those of transformation of the original signal, considered as functions of one discrete argument, into a secondary function of another nature on the same carrier, by way of coordinating the local signal-dependent information and a priori smoothness constraints. In this work, a statistical approach to the generalized edge-preserving smoothing is considered on the basis of treating the sought-for result of processing as a realization of a Markov random process, whose Markov continuity is locally broken at the assumed jump-points. The principal idea of the approach consists in finding the break-points one by one and incorporating them into the model of the hidden process as they are found, so that, at each step, the most detectable of not yet legitimated peculiarities is sought for.