Using innate visual biases to guide face learning in natural scenes: a computational investigation

Area of Science:

• Cognitive Science
• Developmental Psychology
• Computational Neuroscience

Background:

• Newborn infants exhibit an innate bias for orienting to and tracking human faces.
• The underlying mechanism for this face preference is debated, with two main competing theories: the structural and sensory hypotheses.

Purpose of the Study:

• To computationally investigate the effectiveness of the structural and sensory hypotheses in supporting face detection.
• To compare the selectivity of both models in cluttered natural scenes.

Main Methods:

• A computational model was developed to simulate face detection based on the structural hypothesis (a '3-dot' geometric representation).
• A second computational model was developed to simulate face detection based on the sensory hypothesis (generic visual preferences for asymmetry, symmetry, and contrast).
• Both models were evaluated on their ability to detect faces in cluttered natural scenes.

Main Results:

• Both the structural and sensory models demonstrated capability in locating faces within cluttered environments.
• The sensory hypothesis model exhibited greater selectivity in face detection compared to the structural hypothesis model.

Conclusions:

• The sensory hypothesis provides a more selective mechanism for face detection in newborns compared to the structural hypothesis.
• Computational modeling supports the sensory hypothesis as a plausible explanation for infant face preference, particularly in complex visual environments.