Daniel Martins Antunes

We propose a graph cut model to optimize bidimensional shapes with respect to the elastica energy and we show how our iterative and dicrete process relates with the curve-shortening flow. Our model can be included in an image segmentation pipeline and we provide a study to evaluate its effects on the segmentation task.

• Report
• Slides

The thesis investigate image analysis methods based on geometric priors, mainly curvature. The thesis presents an overview of continuous and discrete methods and introduce four new models where we explore the use of multigrid convergent estimators to estimate the geometric quantities that are used as priors.

• Paper
• Slides

Combinatorial model for image segmentation with curvature regularization.

• Paper
• Slides

We prove a tight bound for a local Hall’s theorem for planar graphs. This bound is appliable to the complexity analysis of several combinatorial problems based on local search.

• Paper
• Slides