Optimal disparity estimation in natural stereo images

Area of Science:

• Systems neuroscience
• Computational neuroscience
• Vision science

Background:

• Perceptual constancy is crucial for stable visual perception, allowing recognition of object properties despite variations in sensory input.
• The visual system transforms retinal signals to achieve selectivity for relevant environmental features and invariance to irrelevant image variations.
• Understanding these transformations is a key challenge in systems neuroscience.

Purpose of the Study:

• To develop a method for identifying optimal computations underlying perceptual constancy.
• To apply this method to the specific task of estimating binocular disparity, a critical depth cue.
• To provide a normative account of neural and behavioral aspects of absolute disparity processing.

Main Methods:

• Simultaneously determining optimal receptive field populations for encoding natural stereo images.
• Identifying optimal nonlinear units for decoding population responses into disparity estimates.
• Analyzing photoreceptor responses to natural images to derive processing rules.

Main Results:

• The derived optimal processing rules predict established properties of cortical neurons.
• The computational model's estimation performance parallels key aspects of human psychophysical performance.
• Optimal processing rules are dictated by the statistical properties of natural stimuli, not arbitrary fitting.

Conclusions:

• The study provides a normative framework for understanding how the visual system achieves selective and invariant representations.
• This approach explains neurophysiological and psychophysical findings in absolute disparity processing.
• The findings highlight how neural systems can implement optimal strategies for maximizing performance on specific visual tasks.