Graph Cuts and Approximation of the Euclidean Metric on Anisotropic Grids

Warning

This publication doesn't include Faculty of Economics and Administration. It includes Faculty of Informatics. Official publication website can be found on muni.cz.
Authors

DANĚK Ondřej MATULA Pavel

Year of publication 2010
Type Article in Proceedings
Conference VISAPP International Conference on Computer Vision Theory and Applications
MU Faculty or unit

Faculty of Informatics

Citation
Field Informatics
Keywords graph cuts; euclidean metric approximation; anisotropic grids; voronoi diagrams; image segmentation
Description Graph cuts can be used to find globally minimal contours and surfaces in 2D and 3D space, respectively. To achieve this, weights of the edges in the graph are set so the capacity of the cut approximates the contour length or surface area under chosen metric. Formulas giving good approximation in the case of the Euclidean metric are known, however, they assume isotropic resolution of the underlying grid of pixels or voxels. Anisotropy has to be simulated using more general Riemannian metrics. In this paper we show how to circumvent this and obtain a good approximation of the Euclidean metric on anisotropic grids directly by exploiting the well-known Cauchy-Crofton formulas and Voronoi diagrams theory. Furthermore, we show that our approach yields much smaller metrication errors and most interestingly, it is in particular situations better even in the isotropic case due to its invariance to mirroring. Finally, we demonstrate an application of the derived formulas to biomedical image segmentation.
Related projects:

You are running an old browser version. We recommend updating your browser to its latest version.