Abstract
Faces are thought to be processed holistically, as opposed to through a piecemeal, part-by-part synthesis. This process is so robust that partially occluded faces (e.g., amodally complete behind bars) are nonetheless easy to recognize. This suggests that face recognition may operate despite incomplete visual information (i.e., conceptual representations). Here, we explored whether observers could derive high-level ensemble representations, that is, summary statistical representations, from sets of amodally completing faces. Observers viewed sets of faces varying in identity and adjusted a test face to match the perceived average identity. Faces were linearly interpolated morphs across three identities. The whole set comprised 360 images forming a 'face wheel.' In one condition, the faces amodally completed behind black, horizontal bars. In another condition, the identical facial information was presented, but in the foreground (i.e., the face parts appeared on three-dimensional, fragmented strips in front of a black background). Baseline performance was determined by having observers view and adjust the original, un-occluded faces. The results revealed that the ensemble representation of amodally completing sets was significantly better than the fragmented sets and was not significantly different than the baseline condition. This suggests that high-level face ensembles may be represented conceptually, but that this representation is best when the faces amodally complete.
Meeting abstract presented at VSS 2016