Modeling categorization of scenes containing consistent versus inconsistent objects.

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Computer Vision

Background:

• Previous research indicated faster scene categorization with consistent objects.
• A prevailing hypothesis suggested this advantage arose from rapid object recognition influencing scene perception.
• The role of explicit object recognition versus global scene properties remained unclear.

Purpose of the Study:

• To investigate whether scene categorization can be explained by global scene statistics alone, without explicit object recognition.
• To test a computational model integrating scene statistics and decision-making processes.
• To determine if a simpler mechanism could account for the consistent-object advantage in scene perception.

Main Methods:

• A computational model combining global scene statistics (Oliva & Torralba, 2001) with a diffusion model of perceptual decision making (Ratcliff, 1978) was employed.
• The model simulated ultrarapid scene categorization tasks.
• Model performance was evaluated against experimental findings on consistent-object advantage.

Main Results:

• The integrated computational model successfully accounted for the consistent-object advantage in scene categorization.
• Simulations indicated that the advantage could be explained without invoking ultrarapid object recognition.
• The findings suggest that scene statistics inherently influence categorization performance.

Conclusions:

• The consistent-object advantage in ultrarapid scene categorization may not require direct influence from object recognition.
• Global scene statistics play a crucial role in scene perception and categorization.
• A simpler mechanism based on scene statistics and decision-making models can explain observed behavioral effects.