Lauren Williams¹, Jonathan Flombaum¹, Justin Halberda¹; ¹Johns Hopkins University

Faces are harder to recognize when they appear upside down—an inversion cost. This classic result is explained by a shift from configural & component processing to only component processing. However, this shift in encoding is not well characterized in the face literature through intermediate rotations. Is a face also harder to recognize when it is nearly upside down, e.g. rotated by 135°? How hard is it, comparatively, when the rotation is 90° or 45°? Face recognition as a function of orientation—from upright to upside down—should reveal how and why face processing mechanisms absorb costs for inversion. We therefore investigated face recognition as a function of orientation using flicker change detection. In Experiment 1, participants searched for a changing face within an array of five (drawn from the Chicago Face Database) at orientations between 0° and 180° in 45° steps. When all the faces were upside down reaction times were significantly longer than when all were upright—an inversion cost. Reaction times increased through the intermediate rotations following a sigmoidal function, with an inflection at 90°. These results are consistent with a monotonic, non-linear loss of configural signal as faces rotate from upright to inverted rather than an immediate and complete loss at any non-upright orientation. In Experiment 2, we characterized the inversion cost for other objects (scenes, animals, and vehicles). If faces are special, they should be an outlier among our tested objects. However, all object classes contributed to the positive relationship between time to detect an item change and the inversion effect – i.e., the magnitude of the inversion effect scales with the overall difficulty of identification. That is, all categories show an inversion effect that is proportional to the difficulty of processing that item type, and faces are not special in this regard.