A Morphable Model for the Synthesis of 3D Faces

How to read this

Category

Method: a statistical (morphable) 3D face model

Contributions

• A morphable face model: shape and texture of laser-scanned prototype faces cast into a vector space, with new faces as regulated linear combinations
• An analysis-by-synthesis algorithm that reconstructs 3D shape and texture from one or more photographs by optimizing model and rendering parameters
• A bootstrapping optic-flow procedure for dense correspondence, plus control of attributes like gender, fullness, and distinctiveness

Context

A foundational statistical-shape-model approach for faces, ancestor of essentially all later data-driven 3D face models, drawing on PCA-style example-based modeling rather than a single prior graphics paper.

Correctness

The face space is only as expressive as its laser-scan dataset and the assumption of dense correspondence and linear shape/texture combination; analysis-by-synthesis fitting is an optimization that can depend on initialization and imaging conditions.

Clarity

Dense, with statistics and optimization; a first pass conveys the model concept, a careful second (and third) pass is needed for the correspondence and fitting machinery.

How to read it

First pass for the linear face-space idea and analysis-by-synthesis framing; second/third pass on the correspondence bootstrapping and the fitting optimization if you work with face models.