The Fourth Asian Conference on Vision (ACV2006)i]j

Computation underlying stereopsis depends on the stimulus refresh rate
Maki Takano, Takahiro Doi, Ichiro Fujita
Graduate School of Frontier Biosciences, Osaka University, Osaka, Japan

1. Introduction
The visual system combines signals from two eyes and derives information about the three-dimensional structure of visual scene. To accomplish stereo vision, the visual system must determine which visual objects in one eyefs image correspond to those in the other eyefs image (correspondence problem). In the initial stage of stereo processing neurons compute disparity energy, and do not solve the correspondence problem. Some of the later processing stages retain the disparity energy information, and other stages perform further computation to solve the correspondence problem. The aim of this study was to examine how the disparity energy computation and the correspondence computation underlie stereoscopic depth perception. Especially we investigated the relationship of the two computations to binocular depth perception for visual stimuli with various temporal properties.

2. Methods
Human subjects were engaged in a depth discrimination task with dynamic random-dot stereograms (RDSs). To change temporal properties of visual inputs, we used RDSs with various refresh rates of their dot patterns. In each experiment, we fixed the refresh rate and manipulated the percentage of same-contrast dots between the two eyes (binocularly correlated dots) by reversing the luminance contrast of some dots in one eye (anticorrelation). The neural system for computing disparity energy is sensitive to binocular disparity embedded in anti-correlated RDSs, while the neural system with the correspondence problem solved is insensitive. Thus, the manipulation of the ratio between binocularly correlated and anti-correlated dots enabled us to test the computation underlying stereo depth perception.

3. Results
At fast refresh rates (20-40 Hz), the subjects perceived reversed depth (i.e., near for uncrossed disparity, far for crossed disparity) when anti-correlated dots dominated. Discrimination performance was at chance level when the two types of dots were nearly balanced. The psychometric function as the proportion of correct choices plotted against the binocular correlation level was odd-symmetric with respect to zero correlation (the balanced point). Stereoscopic depth judgments with rapidly changing visual stimuli depended on the correlation level in a manner that negative correlation resulted in reversed depth perception. At slow refresh rates (5-10 Hz), the tendency of perceiving reversed depth for anti-correlated RDSs was weak. The subjects perceived correct depth when the two types of dots were balanced. These results suggest that at slow refresh rates, anti-correlated dots had little effects on depth judgment. The judgment depended solely on correlated dots, which can be used to derive the globally consistent solution of the correspondence problem.

4. Discussion
The results suggest that when visual inputs are rapidly changing, stereoscopic depth perception reflects the energy computation, whereas the correspondence computation underlies depth perception for sustained visual inputs.

Acknowledgments This work was supported by MEXT (17022025) and the Takeda Science Foundation. Correspondence: