The grouping of contours into an L-Vertex depends on contrast polarity: Evidence for the incorporation of image statistics into mechanisms of perceptual grouping

Abstract

The grouping of line segments into a single long contour is independent of changes in contrast polarity. That is, a contour is as readily perceived when some of its elements are lighter and some darker than the background as when all the elements are lighter or darker. This result is consistent with what might be expected if the mechanisms for smooth continuation were shaped by image statistics insofar as it is relatively common for contrast reversals to occur along an extended contour. An L-vertex provides strong evidence for the end of a surface. Would the perception of an L-vertex be invariant to a change in contrast polarity at the point of cotermination of its two segments? If the coding of such vertices has been shaped by image statistics, the perception of L-vertices would be expected to suffer from a polarity change insofar as it would be rare for a change in contrast to occur right at the cotermination point. Subjects named line drawings of common objects, drawn in black or white against a gray background, in which gaps were introduced in the longer contours. Lines were added orthogonal to the object contour at each side of the gap to form either two L-vertices at each gap in some versions of an object or two T-junctions at each gap in other versions. The added lines could be the same direction of contrast as the object's contours at the gap or the opposite contrast. Whereas grouping would be expected to continue through a gap bridged by matching T-junctions, L-vertices would be expected to interfere with the grouping of the lines across the gap, resulting in poorer naming performance. This indeed occurred—when the legs of the L were of the same direction of contrast. When the legs of the L differed in contrast, performance improved to that of the T-junctions (which were unaffected by the direction of contrast), suggesting that the L was no longer effective in defining the end of a surface.