Signal Processing on Textured Meshes

Fabian Prada Nino, Johns Hopkins University

In this thesis we extend signal processing techniques originally formulated in the context of image processing to techniques that can be applied to signals on arbitrary triangles meshes. We develop methods for the two most common representations of signals on triangle meshes: signals sampled at the vertices of a finely tessellated mesh, and signals mapped to a coarsely tessellated mesh through texture maps. Our first contribution is the combination of Lagrangian Integration and the Finite Elements Method in the formulation of two signal processing tasks: Shock Filters for texture and geometry sharpening, and Optical Flow for texture registration. Our second contribution is the formulation of Gradient-Domain processing within the texture atlas. We define a function space that handles chart discontinuities, and linear operators that capture the metric distortion introduced by the parameterization. Our third contribution is the construction of a spatiotemporal atlas parameterization for evolving meshes. Our method introduces localized remeshing operations and a compact parameterization that improves geometry and texture video compression. We show temporally coherent signal processing using partial correspondences.

Speaker Biography

Fabian Prada is a PhD candidate advised by Professor Michael Kazhdan. His main areas of research are Geometry Processing and Geometric Modeling . His research interests spans Computer Graphics, Computer Vision and Scientific Visualization.. Fabian was intern at Microsoft Research in the summers of 2015 and 2016 working on tracking and parameterization of dynamic meshes. Before joining the PhD program, Fabian got a master degree in Mathematics from IMPA (Brazil), and a bachelor degree in Mathematics from Universidad de los Andes (Colombia).