Abstract
Previous research has demonstrated that changes in nonaccidental properties such as cotermination or parallelism are more easily detected than changes in purely metric properties such as object length (Biederman & Bar, 1999). Nonaccidental properties are especially useful for object recognition because they remain relatively stable over different vantage points. However, some of these properties are more stable than others (Chen, 2005), and the present research was designed to investigate whether this has any influence on their relative perceptual salience. Observers performed a match-to-sample task for 2D shapes in which they were required to detect a change in the cotermination of edges (a topological property), a change from straight to curved edges (a projective property), a change from parallel to nonparallel edges (an affine property), or a change in object length (a purely metric property). The results revealed that discrimination thresholds varied systematically among the different types of shape changes. Changes in the pattern of cotermination were the easiest to detect, followed by differences between straight and curved edges, and differences between parallel and nonparallel edges. Purely metric changes in object length produced the highest thresholds of all. We also examined several possible metrics for scaling the differences between object shapes in an effort to determine if any of them are predictive of human performance. These included the correlation or Euclidean distance between pixel intensities or wavelet responses, and a maximum displacement measure originally proposed by Hausdorff. The Hausdorff metric provided the best overall fit to the data, but none of these measures can explain the systematic differences among the different types of shape change.
Meeting abstract presented at VSS 2012