Abstract
PURPOSE: There is a body of literature claiming that “closure” helps strengthen low-level contour grouping. However, it is difficult to know whether closure effects are the result of a low-level closure mechanism, or the result of a low-level “good continuation” mechanism followed by a high-level decision process that picks between grouped contour elements. We attempted to design experiments with better control of decision factors. METHODS: Using a two alternative forced choice procedure, we measured detection accuracy for closed and open contours made up of unconnected line segments. The closed contour was a circle passing through the origin and randomly rotated about the origin; the open contour was a randomly oriented ‘s’ whose center passed through the origin. The ‘s’ stimulus was identical to the circle in almost all aspects except closure. In one experiment, the target contours were embedded in a background consisting of randomly oriented and positioned line segments, and we varied the orientation jitter of the contour line segments. In a second experiment, there were no background elements and we varied the luminance contrast of the line segments. In a third experiment, the background was again randomly oriented line segments and we varied the length of a partial contour presented in the non-target interval. RESULTS: For the first two experiments we found no significant closure effect. This is a departure from what has been found previously and is presumably the consequence of controlling for decision factors. For the third experiment, we found better performance for closed contours. CONCLUSIONS: Closure effects are probably smaller than has been previously thought, requiring specialized conditions unlikely to occur in the natural world. We are exploring contour grouping models (based on natural scene statistics) to determine whether a low-level good continuation mechanism is sufficient to explain the observed closure effect.