Physiological computation of binocular disparity

Area of Science:

• Computational neuroscience
• Vision science

Background:

• A physiologically realistic model for stereo vision was previously proposed.
• This model is based on quantitative binocular receptive field profiles.

Purpose of the Study:

• To present new results that elucidate physiological processes in disparity computation.
• To extend and improve the existing stereo vision model.

Main Methods:

• Extended the model to a more general class of receptive field profiles beyond Gabor functions.
• Demonstrated that Gabor filters offer an advantage, introducing a bias towards zero disparity similar to human perception.
• Proved that complex cells compute disparity via cross-products of filtered retinal image patches, overcoming standard correlator limitations.
• Showed that two complex cells per spatial location suffice for disparity estimation.
• Incorporated larger receptive fields for complex cells by averaging over quadrature pairs of simple cells.
• Evaluated the model using random dot stereograms.

Main Results:

• The model's generality was increased, accommodating various receptive field profiles.
• Gabor filters in the model replicate human perception's bias towards zero disparity.
• A novel computation method for complex cells was identified, improving upon cross-correlation.
• Minimal complex cells (two per location) are sufficient for accurate disparity estimation.
• Averaging simple cell receptive fields significantly improved complex cell reliability and disparity map accuracy.
• Disparity maps generated by the improved model closely resemble human perception, showing sharp transitions.

Conclusions:

• The refined model offers a more robust and physiologically plausible mechanism for stereo vision and disparity computation.
• The model's ability to replicate human perceptual biases and generate accurate disparity maps validates its approach.
• The findings contribute to understanding the neural basis of depth perception.