Abstract
Humans are highly sensitive to incomplete and partial visual information, resulting in object and scene recognition that proceeds seemingly unimpaired by occlusions and noise. Many examples of this have been reported, including contour and surface-based filling-in and patten completion. However, less is known about whether and how pattern completion for high-level object and face representations occurs. Here, we tested for pattern completion of holistic object representations by measuring classification images of Mooney faces. Mooney faces are two-tone black and white blobs that are readily perceived as faces despite lacking low-level face features. In our experiment, participants viewed two identical Mooney face images embedded in random but complementary noise and they selected which was more female-like. Classification images were computed by averaging the selected noise images, and null distributions were generated from shuffled responses. We operationalized pattern completion as the information present in the classification image only within the black regions of the Mooney face. These black regions do not contain any signal, so any consistency in the selected noise images within these areas would suggest that the observer has perceptually completed parts of the face that were not originally there. We found significant within-subject correlations within these black regions, indicating that there is pattern completion in high-level object representations. The results cannot be explained by known contour integration, surface filling-in, or feature-based pattern completion processes. It does, however, corroborate the commonly held assumption that Mooney face recognition is supported by matching the impoverished images to stored templates.