Global versus local internal disparity noise in stereovision

Stereo thresholds are substantially worse for detecting absolute disparity than for relative disparity. One hypothesis is that absolute disparity thresholds are more affected by internal disparity noise than relative disparity thresholds. To test this hypothesis, we measured the equivalent internal disparity noise by adding external disparity noise to random-Gabor-patch (RGP) stereograms, i.e., adding position jitter to paired Gabor patches. For measuring absolute disparity thresholds, the two RGP stereograms with external disparity noise were presented in two temporal intervals, one with mean crossed and the other with mean uncrossed disparity. The task was to detect which interval was closer. For measuring relative disparity thresholds, an RGP stereogram with different disparity polarities (crossed/uncrossed) in the top and bottom halves, was presented with external disparity noise. The task was to detect whether the top or bottom half of the stereogram was closer. We tested 5 spatial frequencies and 6 external noise conditions for both absolute and relative disparity detection. We used the constant stimulus method to measure the minimum and maximum disparity thresholds (Dmin and Dmax). We found that both Dmin and Dmax are substantially lower for relative than for absolute disparity. Dmax remained constant when external noise varied. An equivalent noise model with both global and local internal disparity noise provides a unified explanation of both relative and absolute Dmin thresholds. By assuming that the relative disparity is the difference between absolute disparities, the global internal noise is canceled in relative disparity detection, resulting in a much lower Dmin threshold for relative disparity. Modeling shows that the global internal noise was independent of spatial frequency, while the local internal noise decreased when the spatial frequency increased. Global internal disparity noise may result from vergence noise and/or the absence of conscious readout of absolute disparity, i.e., the absolute disparity anomaly.