Functional MRI Studies on Cross-Correlation and Cross-Matching Binocular Disparity Representations across the Human Visual Cortex

概要

Perceiving binocular disparity-defined depth in visual scenes requires identifying the corresponding points in the left and right retinal images. The initial stage of solving this stereo correspondence problem occurs in V1 that represents binocular disparity cues based on a computation similar to cross-correlation. The represented information still carries binocularly mismatched signals and therefore needs to be transformed into a cross-matching representation in which false match responses are suppressed. It remains elusive where this transformation from cross-correlation to cross-matching representation occurs across the visual areas of the human visual cortex. To address this issue, it requires estimating the proportion of cross-correlation and cross-matching representations encoded across the stages of visual processing. I engineered random-dot stereograms with three dot match levels: anti-correlated, half-matched, and correlated dots. I fitted the brain region’s representation elicited by those stimuli with the linear weighted sum of cross-correlation and cross-matching representational model outputs to get an estimate of the model’s weights. I found that the response patterns in each studied brain region comprise these two representations with different proportions. The areas V1, V2, and V3 had a propensity for cross-correlation representation. The representation in areas V3A, V3B, V7, hV4, and MT was more similar to cross-matching representation. My results suggest that the stereoscopic processing proceeds across visual areas with gradually transforming cross-correlation representation in the early stages toward the cross-matching representation in the higher areas, thus solving the correspondence problem.