A Psychophysical Approach to the Mechanism of Human Stereovision

摘要

Binocular complex cells in primary visual cortex encode disparity based on their spatio-temporal receptive field structure. A probabilistic neural network model based on pre-tuned estimators is proposed as general mechanism of stereovision in V1. Temporal and sapatial features of stereopsis were studied with single image random dot stereograms (SIRDS). Results: Discrimination threshold of two transparent surfaces in a stereogram depends on disparity of each surface as well as the spatial frequency content. Repeated brief exposures (13.4ms) to stereogram against noise or another stereogram do not disturb stereopsis, even when mask intervals are 53ms (P<0.01). With repeated exposures to longer duration (53ms), subjects performed better than expected by temporal averaging (P,0.01). New findings are discussed on the framework of the model. It was concluded that 1. Disparity estimators are arranged in layers according to their preferred spectral and disparity-selectivity, and 2. Stereopsis has two temporal steps: first local estimates are detected (local stereopsis), and ten coherence between different disparity estimators is detected (global stereopsis).