Constant curvature modeling of abstract shape representation

Area of Science:

• Cognitive Science
• Neuroscience
• Computer Vision

Background:

• Understanding abstract shape perception is a key challenge in cognitive science.
• Recent research suggests the visual system encodes contours using constant curvature segments.
• A computational model is needed to explain how boundary points are recoded into this representation.

Purpose of the Study:

• To develop and empirically validate a model of shape representation based on constant curvature segments.
• To investigate the trade-off between representational fidelity and simplicity in visual perception.
• To provide a testable framework for abstract shape encoding in the human visual system.

Main Methods:

• Developed a computational model with two parameters controlling fidelity and simplicity.
• Conducted experiments to empirically estimate model parameters by testing contour discrimination.
• Used the parameterized model to generate shape pairs varying in representational similarity for recognition tasks.

Main Results:

• Identified a clear transition point for discriminating contours based on curvature complexity.
• Empirically determined model parameters reflecting the visual system's balance of fidelity and simplicity.
• Demonstrated that shapes with lower representational similarity (more segments) are easier to discriminate.

Conclusions:

• Provides evidence for constant curvature as a basis for abstract shape representation in human vision.
• The developed model successfully predicts perceptual performance in shape discrimination tasks.
• Offers a testable mechanism for how the visual system encodes abstract shape information.