Some Improvements to Bayesian Image Segmentation Part Two: Classification
Quelques Améliorations à la Segmentation d'Images Bayesienne Seconde Partie: Classification
OPEN ACCESS
We consider the automatic classification framework, using a Markov model . We address the problem of estimating the number of classes and the associated class parameters. We propose a method using the contextual information inherent in images to discriminate different classes in the case of mixture distributions with strongly mixed classes . This method is validated theoretically and practically, using synthetical images and real data . We prove that the proposed method has a validity domain larger than the methods based on a histogram analysis. We then discuss the shape of the data driven potential induced by the detected classes in a Markovian framework. Results are obtained by using two priors :the Potts model and the Chien-model.
Résumé
Nous nous plaçons dans le cadre de la classification automatique. Nous abordons le problème de l'estimation du nombre de classes et des paramètres qui leurs sont associés. Nous proposons une méthode utilisant l'hypothèse contextuelle inhérente aux images pour discriminer les différentes classes . Cette méthode est validée à la fois sur le plan théorique et sur des images de synthèse et des images réelles. Nous montrons, en outre, que la méthode proposée a un domaine de validité plus étendu que les méthodes fondées sur une analyse des modes de l'histogramme. Nous discutons ensuite de la forme du potentiel d'attache aux données dérivé de cette classification dans le cadre d'une segmentation markovienne. Les résultats sont obtenus avec deux modèles a priori différents : le modèle de Potts et le chien-modèle.
Image segmentation, Markov Random Fields, estimation, mixtures .
Mots clés
Segmentation d'images, Champs de Markov, estimation, mélange de distributions.
[I] J. Besag. Spatial Interaction and Statistical Analysis of Lattice Systems. J. Royal Statistical Society, Series B,36 : 721-741, 1974.
[2] J. Besag. On the statistical analysis of dirty pictures . J . Royal Statistical Society, Series B,48 : 259-302, 1986.
[3] M.J. Carlotto. Histogram analysis using a Scale-Space approach.IEEE trans. on PatternAnalysis and Machine Intelligence, 9(l): 121-129, January 1987. 121-129,
[4] B. Chalmond. Image restoration using an estimated Markov model. Signal Processing, 15 : 115-129, 1988.
[5] J. Zerubia, R. Chellappa. Mean field approximation using Compound Gauss Markov Random fields for edge detection and image estimation.IEEE trans. on Neural Networks, 8(4) :703-709, July 1993.
[6] C.S. Won, H. Derin. Unsupervised segmentation of noisy and textured images using Markov Random Fields.Computer Vision Graphics and Image Processing,4: 308-328, 1992.
[7] S. Lakshmanan, H. Derin. Simultaneous Parameter Estimation and Segmentation of Gibbs Random Fields Using Simulated IEEE trans. on Pattern Analysis and Machine Intelligence, 11(8) : 799-813, August 1989.
[8] X. Descombes. A fission and fusion Markovian approach for multi-channel segmentation. InProc. IGARSS'95, Firenze, Italy, pages 124-127, 1995.
[9] H. Derin, H. Elliott. Modelling and Segmentation of Noisy and Textured Images Using Gibbs Random Fields.IEEE trans. on Pattern Analysis and Machine Intelligence, 9 (1) : 39-55, January 1987.
[10] W. Press, S. Teukolski, W. Vetterling, B. Flannery. Numerical Recipes in C The Art of Scientific Computing.Cambridge University Press, 2nd ed., 1992.
[11] S. Geman, D. Geman. Stochastic relaxation, Gibbs distribution, and the Bayesian restoration of images.IEEE trans. on Pattern Analysis and Machine Intelligence, 6 (6):721-741, 1984.
[12] L. Aurdal, X. Descombes, H. Maître, I. Bloch,C. Adamsbaum, G. Kalifa. Fully Automated Analysis of adrenoleukodystrophy from dual echo MRimages. In CAR'95 Berlin,pages 35-40, 1995.
[13] Duda, P.E. Hart.Pattern Classification and Scene Analysis. John Wiley and Sons, New York, 1973.
[14] D.M. Titterington, A.F.M. Smith, U.E. Makov. Statistical Analysis offinite mixture Distributions.John Wiley and Sons, 1985.
[15] A. Goshtasby, W.D. O'Neill. Curve fitting by a sum of Gaussians. CVGIP Graphical Models and Image Processing, 56(4) : 281-288, 1994.
[16] P. Masson, W. Pieczynski, SEM Algorithm and Unsupervised Statistical Segmentation of Satellite Images. IEEE trans. on Geoscience and Remote Sensing 31(3) : 618-633, May 1993.
[17] W. Pieczynski. Champs de Markov cachés et estimation conditionnelle itérative; Traitementdu signal, 11(2), 1994.
[18] Z. Kato, M. Berthod, J. Zerubia, W. Pieczynski, Unsupervised adaptative image segmentation. In proc. ICASSP'95, Detroit, Michigan, USA, pages 2399-2402, May 1995.
[19] M. Sigelle, R. Ronfard. Relaxation of previously classified images by a Markov field technique and its relationship with statistical physics. In Proc. 7th SCIA 91, Aalborg, Danmark,pages 387-394, 1991.
[20] A.P. Dempster, N.M. Laird, D.B. Rubin. Maximum Likelihood from incomplete data via the EM algorithm. J. Royal Statistical Society, Serie B,39 (1) : 1-38, 1977.
[21] C. J. Geyer, E.E. Thompson. Constrained Monte CarloMaximum Likelihood for Dependent Data.J . Royal Statististical Society, serie B, 54(3) : 657-699, 1992.
[22] A.P. Witkin, Scale space filtering :a new approach to multi-scale description. Proc. Image Understanding, pages 79-95, 1984.
[23] X. Descombes, R. Morris, J. Zerubia. Quelques améliorationsà la segmentation d'images bayésienne. Première partie : modélisation. traitement du signal, 1997.