Orientation-position coding and invariance characteristics of pattern discrimination

In this paper, we consider how correlated information about the orientation and position of elements in an image may be used in the detection or discrimination of patterns. Our previously reported experiments (Caelli & Dodwell, 1982; Dodwell & Caelli, in press) demonstrated that discrimination cannot be fully predicted from either source of information on its own. Rather, as we demonstrated, the visual system is better able to recognize and discriminate images when their position and orientation codes are highly related or define images that are invariant under similarity transformations. Here we propose and evaluate mechanisms for performing such orientation and position correlations and invariance coding that allow us to predict the discriminability of patterns. Finally, these mechanisms are shown to predict successfully the ability of observers to recognize patterns when rotated—although they are not sufficient, in themselves, to explain how the actual matching processes may occur in this latter task.