IAM   02674
INSTITUTO ARGENTINO DE MATEMATICA ALBERTO CALDERON
Unidad Ejecutora - UE
congresos y reuniones científicas
Título:
"Complexity-based border detection for textured images,", doi: 10.1109/ICASSP.2010.5495339
Autor/es:
TOMÁS CRIVELLI; AGUSTÍN MAILING; BRUNO CERNUSCHI FRÍAS
Lugar:
Dallas, Texas
Reunión:
Congreso; "2010 IEEE International Conference on Acoustics Speech and Signal Processing", ICASSP'2010; 2010
Institución organizadora:
IEEE
Resumen:
In this paper we address the problem of border detectionfor textured images by exploiting the concept of Bayesiancomplexity minimization as a generalization of the MDLparadigm in the Bayesian context. We aim at determining ifa small window at a certain location corresponds to a singleor several texture classes, which here are modeled as GaussianMarkov Random Fields (GMRF), thereby detecting thepresence of borders. For doing this, a set of possible bordercon􀂿gurations are tested by applying a Bayesian decisionrule that includes two terms: a classical likelihood term relatedto the model 􀂿tting error, and a complexity penalizingterm for the number and size of the window subdivisions ineach con􀂿guration. The latter is derived from the BayesianInformation Criterion (BIC) for non-causal Markov modelsrelated to the MDL decision rule. Experiments on syntheticand real textured images segmentation support the approachwith promising results