Abstract
There is a longstanding debate in the literature about the strategies employed by human observers for achieving object recognition. One popular approach is based primarily on the similarity of an image with a previously stored view (Poggio and Edelman, 1990). When adopting that strategy, recognition performance would be expected to be decreased by any manipulation that alters the 2D projected image of an object, such as changes in viewpoint, lighting or texture. An alternative approach is based instead on a specific set of contour features called non-accidental properties (NAP) that remain relatively stable over many types of image changes (Hummel and Biederman, 1992). The present study used a two interval same-different shape discrimination task to investigate how changes in surface texture affect recognition performance. The shape changes observers were required to detect could involve the alteration of an NAP, or the distortion of a metric property (MP) that had no effect on the NAPs. These two types of shape differences were equally detectable when the objects to be compared both had the same surface texture. That was not the case, however, for object pairs that had different surface textures. In the MP conditions, these texture changes produced significant reductions in recognition performance, but they had no measurable effect in the NAP conditions. These results suggest that observers' recognition strategies may vary depending on the available information. Although a feature-based approach may be preferred for distinguishing objects that have different NAPs, an image based approach may be required for recognizing objects that only differ in their metric properties. Hummel, J., and Biederman, I. (1992). Dynamic binding in a neural network for shape recognition. Psychological Review, 99, 480–517. Poggio, T. and Edelman, S. (1990). A network that learns to recognize three-dimensional objects. Nature, 343, 263–266.
This research was supported by a grant from the National Eye Institute (R01-EY12432).