Configural and featural processing during face perception: A new stimulus set

Goedele Van Belle, Michaël De Smet, Peter De Graef, Luc Van Gool, Karl VerfaillieBehavior 2009 · DOI

Abstract

This paper introduces a standardized set of 60 3D-morphed male faces designed to isolate configural (global shape) from featural (part-based) processing. By leveraging 3D morphable models, researchers successfully decoupled facial features from global head structure. Validation studies confirm that global shape processing is significantly more impaired by inversion than feature processing, providing a robust tool for studying the "Face Inversion Effect" and cognitive disorders like prosopagnosia.

Introduction

A central debate in cognitive psychology concerns whether face perception is primarily "featural" (processing individual parts like eyes and mouth) or "holistic/configural" (processing the spatial relationships and global structure). Historically, studying this has been difficult because experimental stimuli often confounded these two types of changes—for instance, comparing a change in eye color to a change in the distance between eyes. Additionally, extrafacial cues such as hair or makeup often contaminate results. This research presents a "clean" stimulus set where global form and local features are manipulated within the same mathematical framework, removing external biases.

Methods

The researchers utilized a 3D morphable model, trained on the USF HumanID database, to generate 30 original Caucasian male faces. To isolate components, they employed a "subtraction" technique:

  1. Global Form: The 3D shape was iteratively smoothed to remove fine details, leaving only the underlying structure of the skull and soft tissue.
  2. Internal Features: These sharp details (eyes, nose, mouth) were isolated by subtracting the global form from the original 3D model.
  3. The Swap: Features from one face were then mapped onto the global form of another, creating hybrid stimuli that allow for independent testing of featural and configural processing.
Conceptual diagram of the feature-form swap technique.

All stimuli were rendered with standardized skin texture, color, and lighting, and were stripped of hair and clothing to ensure focus remained on facial structure.

Datasets

The resulting dataset consists of 60 unique male face stimuli derived from the USF HumanID 3D database. A key advantage of this set is its versatility; each face is rendered in seven different in-depth orientations, ranging from full profile to frontal views.

The dataset includes faces rendered at seven different in-depth orientations.

The supplemental material includes a total of 839 images, including color-coded "Area of Interest" (AOI) maps designed for eye-tracking research.

Results and Key Insights

The stimulus set was validated through a "same/different" matching task involving 113 participants. Key findings include:

  • Sensitivity ($d'$): Participants successfully detected both featural and configural changes significantly above chance levels.
  • The Inversion Effect: While sensitivity for upright faces averaged $d' = 1.19$, it dropped to $0.50$ when faces were inverted.
  • Configural Sensitivity: Crucially, the performance drop upon inversion was significantly more severe for faces differing only in global form. This confirms that the stimulus set effectively isolates the configural mechanisms most sensitive to orientation.

These results suggest that configural processing involves the entire underlying 3D structure of the head, rather than just the simple 2D spacing between features.

Discussion

While the current set provides a high degree of mathematical rigor and control, it is currently limited to Caucasian male faces. Future work is needed to expand the diversity of the set to include female faces and various ethnicities to mitigate potential "own-race" or "own-gender" biases. Additionally, while the set offers multiple static angles, future iterations could incorporate dynamic 3D rotations. Ultimately, this stimulus set provides a standardized tool for researchers to further explore the nuances of holistic face perception without the confounding variables of traditional stimuli.

Summary updated

BibTeX
@inproceedings{belle2009configural,
  author = {Goedele Van Belle and Michaël De Smet and Peter De Graef and Luc Van Gool and Karl Verfaillie},
  title = {{Configural and featural processing during face perception: A new stimulus set}},
  booktitle = {Behavior},
  year = {2009},
  doi = {10.3758/brm.41.2.279},
}
Figure