|M.Sc Student||Kovnatsky Artiom|
|Subject||Fusion of Geometric and Photometric Information in Non-|
Rigid 3D Shape Retrieval
|Department||Department of Applied Mathematics||Supervisors||Professor Ron Kimmel|
|Assistant Professor Michael M. Bronstein|
In this thesis, we explore the use of the diffusion geometry framework for the fusion of geometric and photometric information in local and global shape descriptors.
The construction of the descriptors is based on the definition of mixed geometric-photometric diffusion process on the shape. For this purpose, we consider shapes as manifolds embedded into a high-dimensional space, where the embedding coordinates account both for geometric and photometric information. This allows defining diffusion processes dependent on geometry and color.
Our experimental results show that employing such descriptors improves shape retrieval performance compared to traditional methods based on purely photometric or geometric information.