The categorical coding of relative orientation in visual object recognition

Abstract

Some theories (e.g., Biederman, 1987, Hummel & Biederman, 1992) propose that in the representation used for visual object recognition, the relative orientation of the parts of an object is coded categorically (e.g., parallel, perpendicular, and oblique). The current research provides the first empirical test of this hypothesis. Two experiments were conducted using two part nonsense objects. For each nonsense object, four versions were created in which the orientationsof the two parts were varied. In one experiment (a classification task), subjects had to decide if two sequentially presented objects were the same ignoring any changes in the orientation of the objects parts. In a second experiment (a physical discrimination task), subjects had to decide if two objects (which would sometimes differ in the relative orientation of the parts) were physically identical. In the classification task, subjects showed better recognition performance when both objects were within a category boundary than when they crossed category boundaries. In the physical discrimination task, subjects showed better recognition performance when objects crossed a category boundary than when they were within a category boundary. The results are consistent with recognition models which posit categorical coding of relative orientation in the representations used for visual object recognition, and further suggest that relative orientation constitutes an instance of categorical perception (Harnad, 1987).