Abstract
Rotating depictions of familiar shapes and objects away from a learnt(‘canonical’) orientation often increases the time required for visual identification (RT), in a nearly linear manner. This has suggested to some that the brain ‘rotates’ the misoriented input image back to the canonical view to support recognition. We dispute this conclusion because (a) it implies prior knowledge of the optimal direction and extent of rotation for an as-yet- unrecognized object, but both backward-masking by pattern and word-picture verification show earlier encoding for identity than for orientation; and (b) these same two procedures show there is an efficient form of recognition that is generally stable from at least 60 to 180-deg of rotation. Moreover, whereas the large progressive effects of orientation on RT are limited to novel views and diminish quickly *after* repetition, our results are obtained with novel (non-repeated) depictions, suggesting the initial RT effect is due to the ‘shock’ value of a newly- misoriented object. We conclude that basic-level recognition of common objects relies on direct matching of features for canonical views, and on view-independent visual cues (such as cross-ratios bound by inflection points) for rotated views. The view-independent route is only prone to effects of unexpected views at a later stage involving verification.