Purchase this article with an account.
Yi D Cheng, Michael J Tarr; How to rob a bank and get away with it: Recognizing disguised faces. Journal of Vision 2003;3(9):834. doi: 10.1167/3.9.834.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Numerous studies have examined an observer's ability to recognize faces across transformations (e.g., inversion, contrast reversal, configural and featural manipulations). However, few researchers have investigated how face recognition is affected by disguises. Here, we investigated observers' ability to recognize disguised faces using an old/new recognition memory paradigm. Three levels of learning and three disguise types (hairstyle, facial hair, glasses) were manipulated. In Exp. 1, faces were disguised by changing a single existing feature, that is, changing the hairstyle, facial hair, or glasses from study to test. We found a main of effect of disguise type, with hairstyle or facial hair producing a similar decrease in recognition accuracy, as compared to glasses, which produced a smaller decrease. Across learning conditions, performance resembled a U-shaped curve with lower recognition memory at intermediate levels of learning. This significant trend suggests that observers shift their strategy for remembering an individual face even over relatively moderate levels of experience. Our speculation is that there is a progression from feature-based representations to “holistic” representations that are less reliant on individual features, and consequently, are more robust to changes such as those that occur in disguises. In Exp. 2, faces were disguised by either adding or deleting a single feature. Here the effect of learning differed depending on whether a given face had a feature added or deleted. With increasing experience, recognition accuracy systematically improved for faces with a feature added, whereas recognition accuracy systematically decreased for faces with a feature deleted. In summary, a disguise per se, the type of disguise, and the direction of the disguise all interact with familiarity. Thus, the underlying representations used in face recognition are highly dynamic and may be more or less robust to disguise manipulations.
This PDF is available to Subscribers Only