Recognition of Vector Patterns under Transformations: Local and Global Determinants

Abstract

The extent to which combined local position and orientation information contributes to the recognition of patterns under transformation was investigated. Vector patterns, which consist of arrays of line segments composed according to specific rules, were presented in pairs. Discrimination performance was measured both as a function of the degree of local perturbation in one member of an otherwise identical pair, and in terms of a global rotation of one pattern with respect to the other. Differences were found between vector-pattern types on this task that point to a two-component process for pattern recognition under transformation. One component involves the comparison of local orientation/position information in the original pattern with that in its transform. The second component is global and is related to the degree to which the vector pattern is invariant under certain whole-field (1-parameter) transformations.