Project
Scale-consistent edge-aware regularization
Analysis of discrete weighted energies under mesh refinement and design of resolution-robust anisotropic regularizers.
The project addresses a subtle failure mode of edge-aware regularization: weights built from raw intensity jumps collapse toward isotropy as the grid is refined.
By reformulating the weights through normalized discrete directional derivatives, the project restores dimensional consistency and connects discrete variational problems with meaningful continuous anisotropic limits.
Research areas
Scale-consistent variational regularization
Participants
Evgeny Yuryevich Shchetinin
Andrey Andreyevich Shevchuk
Related publications
On the Well-Posedness of Discrete Variational Problems with Edge-Aware Regularization
A Computational Method for Image Segmentation Based on a Dirichlet Field and an Analysis of the Asymptotic Accuracy of Spatial Regularizer Discretization