Dissociating the effects of viewpoint disparity and image similarity in mental rotation and object recognition.

Abstract

Mental rotation (MR) is often measured in a task where participants judge the handedness of rotated objects, and is revealed by a viewpoint cost - a linear reduction in performance with an increase of viewpoint disparity for the objects. Similar viewpoint costs are also often found for object recognition (OR), in tasks where participants match the identity of rotated objects. These findings led to speculations that MR might be the mechanism underlying OR. However, recent studies dissociate the two tasks (Gauthier et al., 2002; Hayward et al., 2006). If viewpoint costs are dissociable in MR and OR, we may ask whether they arise because of the same or different factors. Specifically, viewpoint disparity is usually confounded by image similarity in these tasks - images become dissimilar with an increase in viewpoint difference. Here, we investigated the effects of viewpoint disparity and image similarity in MR and OR using novel objects rotated around the vertical axis. Subjective similarity ratings for image pairs that differed by 40˚, 80˚, 120˚ and 160˚ were collected and used to dissociate the two factors. In a Similarity condition, the viewpoint disparity was fixed and image similarity was manipulated. In a Viewpoint condition, viewpoint disparities were manipulated while similarity was fixed. In a sequential matching paradigm, participants performed the MR or OR tasks in both Similarity and Viewpoint conditions. Performance was better for similar than dissimilar image pairs for both tasks, but the effect was larger for OR than MR. In contrast, when similarity was controlled, a viewpoint cost was only found for MR but not for OR. These results demonstrate different causes of the viewpoint costs in the two tasks: while MR largely relies on 3-D mental transformation procedures that depend on viewpoint disparity, OR is based predominantly on matching similarity of image features. Acknowledgement: This work was supported by grants from the Research Grants Council of the Hong Kong Special Administrative Region (China), NSF and the James S. McDonnell Foundation.