Selim Esedoglu
Title: TVL1 models for imaging: global optimization and geometric properties: Part II
Abstract: We first discuss how the convex formulation of two-phase,
piecewise constant image segmentation motivated by the L1 fidelity version
of the ROF model allows adapting the convex duality based optimization
methods developed originally for the standard ROF model to the context of
image segmentation. We also briefly indicate recent work on applying these
ideas to other related shape optimization problems. Then, we shift focus and
discuss the correct way to generalize L1 regularization to the context of
geometry denoising, which is a standard problem (also called surface
fairing) in computer graphics.