Abstract
The Spatiotemporal Contrast Sensitivity Surface (STCSS) is a comprehensive characteristic of visual sensitivity (1/threshold), measured using luminance gratings of variable spatial and temporal frequencies of modulation (Kelly, 1979). Estimating the entire STCSS is important for basic and clinical vision research, but it requires massive data collection. Novel adaptive procedures allow one to rapidly estimate the full STCSS (Lesmes et al., 2009) by making assumptions about the form of the STCSS and its underlying psychometric functions. By doing so, estimation at one stimulus condition informs and improves estimation at other conditions. The precision of estimating the STCSS by the adaptive procedures depends on the validity of these assumptions. We explored how different parametric descriptions affect STCSS estimation in different tasks.
Observers discriminated speed or direction of motion in drifting sinusoidal gratings, while grating contrast was varied according to one of two Bayesian adaptive procedures: Psi (Kontsevich and Tyler, 1999) or qCSF (Lesmes et al., 2008). Both procedures use a strategy that selects stimuli expected to yield the largest information gain about psychometric parameters. Psi procedure estimates thresholds and slopes of psychometric functions separately for each spatiotemporal frequency. In speed discrimination, observers reported which stimulus moved faster in a 2-IFC task. In direction discrimination, observers reported which direction they saw among two opposite directions, in a 2-AFC task.
In both tasks we obtained sensitivity surfaces of similar shapes, which supports the view that the STCSS has a task-independent shape (Nakayama, 1986; Gepshtein et al., 2007). The slopes of psychometric functions covaried with thresholds in speed discrimination, but they did not vary in direction discrimination. These results indicate that a complete characterization of visual sensitivity should include both slopes and thresholds of psychometric functions across spatiotemporal frequencies, rather than only the thresholds used in previous parametrizations.