Abstract
Imagine vertically compressing a facial image to half its original width. This grotesquely distorted face would be expected to be much less recognizable relative to the original given the marked changes in facial configuration the compression introduces. We have found, however, that even extreme compressions, leading to sliver-like faces that are just one-sixth the original width or height, leave identification performance almost entirely unaffected. This result has important implications for the nature of configural information that participates in recognition. Although there is currently no precise definition of ‘facial configuration’, it is implicitly assumed that its primary constituents are the mutual spatial relationships of the internal features (eyes, nose and mouth). Here, we report experiments designed to characterize how the relative contributions of internal and external features (hair and jaw-line) change as a function of image compression. We find that even at moderate compressions, the absolute sum of recognition performance with internal features alone on the one hand and external features alone on the other, is significantly lower than performance based on internal and external features together. This highly non-linear relationship suggests that the definition of ‘facial configuration’ must crucially include spatial relationships linking internal and external features rather than merely the mutual arrangement of internal features alone. More generally, recognizing whole faces appears to involve non-linear cue-fusion for integrating internal and external features, rather than being the sum of two independent processes, one for internal features and the other for external ones.