- Neuroscientists have reverse engineered the information that characterizes an individual’s facial identity and rendered it graphically.
- They have built 3D facial models using key details stored in an individual’s brain.
Every time you step out of your home, you’re exposed to dozens of faces: some familiar, some entirely new. Yet with just a glance, your brain analyzes the features of those faces and stores them in memory.
Familiar faces (such as family members, friends and colleagues) are stored in memory with sufficiently versatile details and thus it’s easy to accurately recognize them across diverse common tasks — for instance, identifying your friends in different poses, at different ages.
Now, researchers at the University of Glasgow have reverse engineered the information that characterizes an individual’s identity to mathematically represent it and then render it graphically. They have been able to develop 3D facial models using key details stored in an individual’s brain when recalling a familiar face.
How Did They Do It?
The research team studied how their own work colleagues recognize the faces of other colleagues from memory, and tried to model the 3D face-identity information stored in their memory.
They built an approach based on reverse correlation and a generative model of 3D face identity (GMF), separately for 2D texture information and 3D shape.
The GMF synthesized a set of 6 new 3D faces on every experimental trial. Each face had a unique identity and shared other categorical face information (such as sex, age, and ethnicity) with 1 of the 4 faces personally familiar among colleagues.
To implement this, they used a general linear model to decompose the familiar target face into a module that determines the particular identity of the familiar face — for example, for ‘John’, the average of all black male faces of 40 years of age.
Diagnostic and nondiagnostic components of faithful 3D shape representation of target familiar face | Courtesy of researchers
They then added a new random component to the target face. After seeing all these faces, participants selected the one that most resembled the familiar target.
Researchers carried out several trials, and then estimated the information content of the mental representation of each target face in each participant, using reverse correlation.
In the next phase, they tested with new participants whether these representations were sufficiently detailed and versatile to enable identification and resemblance of each target familiar face across various factors (for instance across siblings, age, and new viewpoints).
They demonstrated the efficacy of the modeled contents in several resemblance tasks. The study will help scientists better understand the brain mechanisms of face identification, and could have applications for artificial intelligence and gaming technology.