EE Seminar: Parameterization Driven 3D Surface Modeling from a Single Image

~~
Speaker: Michael Litvin,
M.Sc. student under the supervision of Prof. Ariel Shamir and Prof. Alex Bronstein

Wednesday, January 6th, 2016 at 15:30
Room 011, Kitot Bldg., Faculty of Engineering

Parameterization Driven 3D Surface Modeling from a Single Image

Abstract

We consider the task of retexturing, which is the process of realistically replacing the texture of an object in an image. The user provides a regular 2D image along with an alternative texture, and uses our system to coarsely model the object’s visible 3D geometry. A parameterization implicitly maintained throughout the modeling process induces a warp that fits the texture on the desired object.
In order to lessen required user interaction and skill, we limit the geometry to extrusion surfaces. Most user input is performed by marking simple geometrical features over the 2D image, and entering optional real-world measurements. The process begins with defining an extrusion spine in form of a line segment, and then defining the shape of key cross-sectional profiles along this spine. The profiles are then automatically scaled to fit the object’s silhouette, sampled and connected to form a 3D triangle mesh. This construction induces a natural parameterization with axes based on the spine direction and the profiles’ arc-length parameter.