Abstract
Contrast detection thresholds improve with stimulus extent. This is often explained by statistical summation between independent spatial channels. Here we thought to test this hypothesis by comparing contrast detection thresholds for different configurations of Gabor quintets. Stimulus parameters were similar to those defined by the ModelFest group (http://vision.arc.nasa.gov/modelfest/) for Gabor Strings. These strings were generated from five horizontal collinear Gabor patches (8 cpd) with 0.35° inter-patch distance, arranged horizontally (= = = = =). In our experiments, there were also strings with parallel Gabor patches (|| || || || ||) and strings with alternating orientations (|| = || = || and = || = || =). Thresholds were measured using a standard staircase procedure with a 2AFC method. Stimuli were presented within a Gaussian temporal envelope with σ=120 or 60 ms. Results from 4 naïve observers showed a significant improvement in threshold for all configurations relative to the threshold of the corresponding central patch. The collinear configuration showed the largest improvement, 0.17 log-units, in agreement with the 4th root summation law [Ct(n)=Ct(1)/n1/4, here n=5], while the other 3 configurations improved by only 0.11 log-units. Stimulus duration had only a small effect on the collinear improvement (0.01 log-units difference) but a somewhat significant effect on the improvement of the non-collinear configurations (0.13 vs 0.08 for 120 and 60 ms respectively, p=0.05, paired t-test). The results do not agree with probability-summation across space as the assumption of channel independence predicts equal improvements in threshold for all configurations. It is possible that equality can be achieved with longer stimulus durations. The results with the shorter duration point to the existence of a neuronal mechanism that is specialized in detecting low contrast contours (Usher et al 1999, Spatial Vision 12, 187–210).