Segmentation as a bottleneck in numerical cognition

Area of Science:

• Cognitive Psychology
• Visual Perception
• Numerical Cognition

Background:

• Ecological relevance of number perception is often assumed but rarely tested with naturalistic stimuli.
• Traditional studies use simplified stimuli (e.g., dot arrays), limiting understanding of real-world number processing.

Purpose of the Study:

• To investigate numerical discrimination in naturalistic visual scenes.
• To compare performance with traditional stimuli (dot arrays) and assess the impact of complexity and occlusion.

Main Methods:

• Adult participants performed number discrimination tasks using four stimulus types: dot arrays, noisy dot arrays, naturalistic images, and pseudo-naturalistic images.
• Stimuli varied in object count, complexity, and occlusion levels.

Main Results:

• Numerical discrimination accuracy was significantly lower for naturalistic and pseudo-naturalistic images compared to dot arrays.
• Performance decline in naturalistic conditions correlated with increased object count and was driven by object occlusion.
• An interaction revealed that object count only affected performance when occlusion was present.

Conclusions:

• Object segmentation, especially with occlusion, creates a bottleneck for numerical discrimination in complex scenes.
• Current models of numerical cognition need to incorporate the computational costs of scene parsing.
• Findings highlight the ecological limitations of number perception in cluttered, real-world environments.