Seminar for Probability Logic



Plan rada Seminara Verovatnosnih logika za decembar 2012.

Sastanci seminara verovatnosnih logika Matematickog instituta SANU odrzavaju se na adresi: Kneza Mihaila 36/III sprat, soba 301f - sala za seminare. Sastanci se odrzavaju cetvrtkom posle podne, od 15h, pod rukovodstvom Profesora Miodraga Raskovica.

ČETVRTAK, 06.12.2012. U 15:15
dr Radosav Raka Djordjevic

ČETVRTAK, 13.12.2012. U 16:00
Natasa Sladoje, Faculty of Technical Sciences, University of Novi Sad, Serbia
Abstract: The coverage model provides a framework for representing continuous objects present in digital images as spatial fuzzy subsets. We introduce the basic definitions and properties of this model and show how it can be used to improve information extraction from digital images and to reduce problems originating from limited spatial resolution. The focus of the talk is on one method for coverage segmentation, where the coverage percentage of each image element (pixel) by each of the image components is estimated. The method enhances a model for linear unmixing of image intensities by incorporating criteria for regulating spatial smoothness and for imposing a theoretically plausible result. The segmentation task is formulated as an optimization problem and is solved by the Spectral Projected Gradient method. Excellent performance of the method in noisy conditions is demonstrated, with a particular emphasis on its combination with feature estimation methods utilizing the coverage information to provide high precision results. The generality of the model, combined with the flexible energy minimization approach, facilitates further extensions; such as the option of super-resolution segmentation as well as processing of blurred signals.

ČETVRTAK, 13.12.2012. U 15:15
Joakim Lindblad, Faculty of Technical Sciences, University of Novi Sad, Serbia and Centre for Image Analysis, Uppsala University, Sweden
DISTANCES BETWEEN FUZZY SETS AND THEIR APPLICATIONS TO IMAGE PROCESSING Abstract: Several studies indicate clear advantages in terms of increased precision and robustness of various image analysis tasks, if fuzzy spatial sets are used for object representations, instead of crisp ones. It is therefore of interest to further develop image processing tools applicable to fuzzy object representations. Distance measures and distance transforms are among the earliest image processing tools, used for image segmentation, object matching and pattern recognition, image registration, object skeletonization, image retrieval, etc. This talk will present two novel path-based point-to-set distance measures and a class of set-to-set distances based on them, which are applicable to fuzzy objects. The newly proposed distances have a number of appealing theoretical properties, in turn making them very appropriate for practical use. The distances are of linear computational complexity which enables their use for addressing large scale problems. Performance evaluation of the set distances on real image processing tasks confirms their excellent performance in pattern matching (generalized to texture analysis) and image registration tasks. This holds well for fuzzy segmented images but also for grey level images directly where they provide an elegant combination of matching both shape and texture. We demonstrate the usefulness of proposed distances for recognition of hand written numbers as well as for finding virus particles in electron microscopy images.

ČETVRTAK, 20.12.2012. U 15:00
Aleksandar Perovic

ČETVRTAK, 27.12.2012. U 15:00
Aleksandar Perovic
Predavanje ce biti odrzano u biblioteci MI SANU.

Beograd, 2012.

Sekretar Seminara: Nataša Glišović
Rukovodilac Seminara: Prof. dr Miodrag Rašković