Cross-matching: a modified cross-correlation underlying threshold energy model and match-based depth perception.

Area of Science:

• Computational Neuroscience
• Visual Perception
• Image Processing

Background:

• Stereoscopic depth perception relies on matching images from the left and right eyes, a process complicated by visual ambiguity.
• The stereo correspondence problem is particularly challenging with random-dot stereograms (RDSs) due to identical dot appearances.
• Anticorrelated RDSs (aRDSs) present a unique challenge as they lack a coherent solution for the correspondence problem.

Purpose of the Study:

• To clarify the essential computation underlying the threshold energy model for visual cortex neurons.
• To propose and validate a new computational model, termed 'cross-matching,' for stereoscopic depth perception.

Main Methods:

• Introduced half-wave rectification into the cross-correlation of left and right eye images, creating the 'cross-matching' computation.
• Simulated psychometric curves using graded anticorrelation to model human near/far discrimination.
• Analyzed disparity tuning derived from cross-matching for both normal and anticorrelated RDSs.

Main Results:

• Cross-matching computation demonstrated attenuated disparity tuning for anticorrelated RDSs (aRDSs).
• Simulated psychometric curves accurately reproduced human depth perception data for varying levels of anticorrelation.
• Predicted decreased performance for aRDSs with increased dot density and a lower anticorrelation threshold for nullifying depth perception.

Conclusions:

• Cross-matching provides a plausible computational mechanism underlying the neural processing of stereoscopic depth.
• This model effectively explains how the visual system resolves the stereo correspondence problem, even with challenging stimuli like aRDSs.
• The findings suggest cross-matching is a fundamental computation for match-based disparity signals in human depth perception.