Visual numerosity perception shows no advantage in real-world scenes compared to artificial displays

Area of Science:

• Cognitive Science
• Neuroscience
• Computer Vision

Background:

• Human visual system is sensitive to numerosity, but underlying mechanisms are unknown.
• Current theories suggest numerosity perception develops from passive exposure to visual scenes.
• Unsupervised deep neural networks are proposed to model behavioral features of numerosity perception.

Purpose of the Study:

• To test if natural scene statistics enhance numerosity perception.
• To challenge experiential learning models of visual number sense.
• To investigate the role of abstract number representation.

Main Methods:

• College-aged adults viewed natural and artificial stimuli.
• Number perception was assessed for both stimulus types.
• Factors like object identification, clutter, and occlusion were considered.

Main Results:

• No advantage, and sometimes impairment, in number perception for natural scenes compared to artificial displays.
• Findings were independent of object identification difficulty, visual clutter, parsability, or occlusion.
• Results contradict predictions from statistical learning and experiential models.

Conclusions:

• Visual number sense is attuned to abstract object counts, not real-world statistical correlations.
• Experiential learning models based on statistical regularities are challenged.
• Object complexity and entropy's role in number perception warrants further investigation.