Perceiving animacy from shape.

Area of Science:

• Cognitive Psychology
• Computer Vision
• Perceptual Science

Background:

• Superordinate visual classification, grouping diverse items (e.g., octopus, dog) into a single class (animal), is computationally complex.
• Learning superordinate categories may involve acquiring generalizable features shared by class members, facilitating novel item classification.

Purpose of the Study:

• To investigate the specific visual shape features associated with superordinate categories, particularly animate and inanimate classes.
• To determine if these shape features can predict category membership for unfamiliar objects.

Main Methods:

• Participants rated unfamiliar items on their resemblance to 'animal,' 'plant,' and 'mineral' categories.
• A second group rated the same items based on 22 distinct midlevel shape features (e.g., symmetry, curvature).
• Linear discriminant analysis was used to assess the predictive power of shape features for perceived categories.

Main Results:

• Systematic tradeoffs in category ratings indicated an inherent class-like organization of the items.
• Specific midlevel shape features were strongly associated with superordinate classes (e.g., 'animals' rated higher on 'symmetry').
• Shape feature vectors accurately predicted perceived categories, comparable to ground truth classification.

Conclusions:

• A common set of midlevel visual shape features underpins superordinate classification along the animacy continuum.
• These findings suggest that general shape characteristics are crucial for categorizing novel and unfamiliar objects.