Variable multilinear models for facial synthesis
US7133048B2 · kind B2 · utility
Assignee
Inventor
Key dates
| Filing date | Jun 30, 2004 |
| Grant date | Nov 7, 2006 |
| Priority date | — |
| Expiry date | Apr 23, 2025 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG06T13/40
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
A method constructs a variable multilinear model representing a class of deformable surfaces. First, meshes of deformable surfaces are acquired. The meshes include vertices. The meshes have different identities and different expressions. The meshes can be obtained from images of human faces, where facial features, such as eyes, eyebrows, cheeks, nose, mouth and chin, form the deformable surfaces. The meshes are stored in a memory as elements of a data tensor. The data tensor is selectively flattened to matrices composed of column vectors. An imputative incremental singular value decomposition is applied to each matrix to generate a set of orthogonal bases. Then, the orthogonal bases are applied to the data tensor, via tensor multiplication, to construct a core tensor, which is the variable multilinear model representing the class of surfaces.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.