Facial Action Coding System

How to read this

Category

psychology coding manual, not a graphics paper

Contributions

• Defines the Facial Action Coding System, a catalog of Action Units, each tied to the contraction of a specific facial muscle or muscle group, that lets a trained human coder decompose any observed expression into an objective combination of those units
• Separates anatomically grounded Action Units from purely descriptive ones, giving facial animation a muscle referenced vocabulary that is independent of any particular emotion label
• Provides a training and scoring methodology for human coders, which is what made FACS reproducible as a measurement tool across labs and studios rather than a one off private taxonomy

Context

FACS predates digital facial animation by years, but the field adopted it wholesale once parametric and blendshape rigs, descending from parke-facial-animation-1974, needed a shared naming scheme for expression controls. It now underlies most production facial rigs' shape and control naming, AU named blendshapes are standard practice, and it underlies most automated expression tracking and performance capture systems, including the personalized FACS based rigs used in andrus-facelab-2020, that need a fixed label set to regress or drive against.

Correctness

This is a manual for a measurement procedure, not a testable scientific claim, so correctness here means procedural validity: it rests on decades of inter rater reliability studies showing trained coders agree on scoring. Its limitation for graphics use is that it was built to measure human expression for psychology research, not to specify a rig topology or blendshape count, so studios still have to decide independently which AUs to expose as controls and how to blend them.

Clarity

A reference manual meant to be consulted, not read cover to cover, dense with muscle anatomy and human coder scoring criteria. Most riggers and TDs today only ever touch the AU number to muscle mapping table rather than the full training material.

How to read it

Do not attempt a linear read. Pull the AU to muscle group table, widely reproduced in summaries and on the Paul Ekman Group site, as the actual working reference. Skim the introduction only if curious how the coding methodology was validated. Keep it as a lookup table to consult alongside a rig's blendshape list, not as study material.