Computational study of depth completion consistent with human bi-stable perception for ambiguous figures

Area of Science:

• Computational neuroscience
• Visual perception
• Psychophysics

Background:

• Human depth perception in regions lacking binocular disparity (ambiguous regions) remains a challenge.
• Existing models struggle with depth calculation in textureless areas where binocular disparity is absent.

Purpose of the Study:

• To propose a computational model consistent with human depth perception in ambiguous visual regions.
• To investigate the influence of prior stimuli on depth perception in these regions.
• To validate model predictions through psychological experiments.

Main Methods:

• Development of a computational model for depth completion in ambiguous regions.
• Computer simulations to generate depth maps and compare with human perception.
• Psychological experiments using ambiguous and prior stimuli with varying inter-stimulus intervals (ISI).

Main Results:

• The computational model qualitatively reproduces human depth perception in ambiguous regions.
• Depth perception in ambiguous regions is dependent on initial depth conditions influenced by prior stimuli.
• Perceptual correlation between prior and post-stimuli varied with ISI duration: positive for short, negative for medium, and near zero for long durations.

Conclusions:

• Human depth perception in ambiguous regions is influenced by preceding visual information.
• A computational model can explain the temporal dynamics of this influence based on ISI.
• The findings offer new insights into the mechanisms of depth completion and surface perception.