The Role of Visual Field Position in Object Recognition

Visual recognition of objects is generally assumed to be independent of the location in the field of view. Experiments have shown, however, that for stimuli such as random-dot clouds recognition can be severely affected by retinal displacement (Foster and Kahn, 1985 Biological Cybernetics51 305 – 312; Nazir and O'Regan, 1990 Spatial Vision5 81 – 100; Dill and Fahle, Perception & Psychophysics in press). In a series of new experiments, we tested whether similar shortcomings of translation invariance can be obtained also with more natural-looking objects. For that purpose, we tested human subjects with 3-D animal-like shapes that had been employed previously in studies of rotation invariance (Edelman, 1995 Biological Cybernetics72 207 – 220). Some of our experiments included same - different discrimination, while in others the subjects had to label the briefly displayed stimulus with one of two possible labels. In both tasks, translation invariance was found to be incomplete: performance was significantly reduced when object memory had to be transferred to new locations. This positional specificity parallels the imperfect generalisation of recognition over rotation in depth, reported in the past years by many research groups. Similar to those findings, our present results suggest that the mechanisms of visual object recognition may be view-based rather than object-based. As before, these results have implications concerning the various theoretical approaches to the understanding of recognition currently under consideration.