Critical Features for Face Recognition

Abstract

Face recognition is a computationally challenging task that humans perform effortlessly. Nonetheless, this remarkable ability is limited to familiar faces and does not generalize to unfamiliar faces. To account for humans’ superior ability to recognize familiar faces, current theories suggest that familiar and unfamiliar faces have different perceptual representations. In the current study, we applied a reverse engineering approach to reveal which facial features are critical for familiar face recognition. In contrast to current views, we discovered that the same subset of features that are used for matching unfamiliar faces, are also used for matching as well as recognition of familiar faces. We further show that these features are also used by a deep neural network face recognition algorithm. We therefore propose a new framework that assumes similar perceptual representation for all faces and integrates cognition and perception to account for humans’ superior recognition of familiar faces.

Decomposing the composite face effect: Evidence for non-holistic processing based on the ex-Gaussian distribution

“…The current findings showed that the composite face effect with familiar and unfamiliar faces has been manifested at the same ex-Gaussian component—the τ component. This common pattern increases support in the hypothesis that familiar and unfamiliar faces are governed by the same mechanisms (Abudarham et al, 2019). However, this conclusion should be constrained by the possibility that different mechanisms can still produce similar influence on the τ component.…”

“…This also strengthens the position that familiar and unfamiliar composites are governed by similar (holistic?) processes/constructs (Abudarham et al, 2019). Moreover, since the τ component is associated with attention and working memory processes (Schmiedek et al, 2007) and/or object-based attention processes (Spieler et al, 2001), the current findings support attentional and object-based accounts of the composite face effect (Fitousi, 2015).…”

“…And their operational definitions have become a matter of debate (Richler & Gauthier, 2014; Rossion, 2013). In addition, researchers have started questioning the idea that familiar and unfamiliar faces are governed by the same holistic mechanisms (Burton et al, 2015; but see Abudarham et al, 2019). The lack of consensus among researchers over these basic issues impedes progress in the field of face perception (Burton, 2013).…”

Conceptual rather than perceptual similarity enables generalization across perceptually different appearances of familiarized faces

“…Here we propose that the view-invariant perceptual representation is generated for all faces regardless of their familiarity. This suggestion is based on our recent findings that show that the same view-invariant critical features are used for both familiar and unfamiliar faces (Abudarham et al, 2019). We therefore propose that the significant gap in performance between familiar and unfamiliar faces is mediated by the conceptual representation of the familiar face, which may correspond to Bruce and Young's face recognition unit (FRU), which unlike the structural code, is not purely perceptual, but links the perceptual and the conceptual knowledge about the familiar person.…”

“…We therefore named them critical features, as they are critical for the identity of a face. In contrast, changing another set of facial features did not change the identity of the face, and we therefore named them non-critical features (Abudarham, Shkiller, & Yovel, 2019;Abudarham & Yovel, 2016 (Figure 3A and Figure S1 and S2). In the learning phase, we presented pairs of images of the same identity that differed in critical features or in non-critical features.…”