Speaker: Mathias Oehmen
Abstract
In this presentation, I will introduce myself as an Erasmus student from the University of Münster and outline my current research project at the Institute of Visual Computing and Human-Centered Technology (PRIP). The project addresses the challenge of reliably separating inliers from outliers in point clouds, as noise often deteriorates the quality of classification and model optimization. To this end, I am investigating a "marriage" of two concepts: the "Self-coNsistent Agreement Principle" (SNAP) developed by Xiaoyi Jiang and Andreas Nienkötter, and the "Contrast Pyramids" established by Walter G. Kropatsch.
The objective is to leverage the agreement-based weights generated by SNAP to guide the contraction kernels of the pyramid, ensuring that structurally consistent information is preserved while outliers are efficiently suppressed. Furthermore, I will present initial considerations regarding the underlying graph topology, comparing options such as Voronoi diagrams and Delaunay triangulations, to establish a robust basis for the hierarchical reduction of point clouds. This talk serves as an introduction to my Erasmus program and the roadmap toward a joint publication for the S+SSPR conference in Bern.
Bio
Mathias Oehmen is an ERASMUS student working with Prof. Walter Kropatsch.