It is well known that faces are difficult to identify in photographic negatives. Exploring the causes of this phenomenon is important for understanding the broader issue of the nature of information the visual system uses for face identification. A prominent explanation for this recognition impairment is that the unnatural shading in negated images compromises the ability of shape-from-shading processes to provide correct 3D facial shape information. We suggest an alternative account that is based on the use of stable 2D contrast relationships. Based on our earlier work showing that contrast polarity around the eyes is stable across various imaging conditions, we hypothesized that contrast negation compromises performance by destroying these highly consistent polarity relations. To test this hypothesis, we created a set of ‘contrast chimeras’. These are faces which are photo-negatives everywhere except in the eye region (thus preserving local contrast polarity in that neighborhood). Given that such faces have unnatural shading cues over much of their extents, a shape-from-shading based explanation would not predict significant improvements in recognition performance. However, if performance in negatives is compromised, at least in part, due to the destruction of local polarity relations between the eyes and their neighborhood, performance with contrast chimeras is expected to be better than that with contrast negatives. We found that contrast chimeras were recognized at a significantly higher rate than the full negatives and the increase in performance could not be explained simply by appealing to the recognizability of the eyes. We infer that ordinal brightness relationships between a few regions of the face are major determinants of facial recognizability, and that destruction of these relationships in negatives may underlie the observed recognition impairments. We also discuss the broader implications of these results.
This research was supported by the DARPA HumanID Program and the Alfred P. Sloan Foundation.