Abstract
Visual perception in a natural environment requires integration of information across space. Previous spatial summation experiments (contrast threshold vs. signal area) with gratings showed that optimal integration occurred only over a limited spatial extent (Kersten, 1984). We measured threshold contrast as a function of signal area in a scene-identification task. The scenes were either masked with correlated Gaussian noise that had an amplitude spectrum similar to those of the scenes, or were presented without noise. An observer's task was to determine if two simultaneously presented pictures (3.5°(w) × 9.9° (h)) were taken from the same scenery (the pictures themselves were always different). Signal area was manipulated by cropping the pictures horizontally from full height to 15% of the full height. In the noise condition (noise RMS contrast = 10%), threshold contrast decreased with signal area with a log-log slope of −1/2 (0.54±0.07) over the entire range of signal area without reaching any plateau. This result suggests that the visual system can optimally integrate task-relevant information over an unlimited spatial extent! In the noise-free condition, threshold contrast vs. signal area had a log-log slope steeper than −1/2 (−0.94±0.18). This result can be explained by postulating that both local and long-range feature detectors mediate scene perception, with the long-range feature detectors being less noise-tolerant and less responsive to partial features. Reducing the signal area leads to more partial features and inactivates disproportionably many long-range feature detectors. The result is a sharp rise in contrast threshold for small signal area and thus a steep summation slope.
Supported by: NIH EY016391