How face perception unfolds over time

Abstract

Within a fraction of a second of viewing a face, we have already determined its gender, age and identity. A full understanding of this remarkable feat will require a characterization of the computational steps it entails, along with the representations extracted at each. Here we used magnetencephalography to ask which properties of a face are extracted when, and how early in processing these computations are affected by face familiarity. Subjects viewed images of familiar and unfamiliar faces varying orthogonally in gender and age. Using representational similarity analysis, we found that gender and age information emerged significantly earlier than identity information, followed by a late signature of familiarity. Importantly, gender and identity representations were enhanced for familiar faces early during processing. These findings start to reveal the sequence of processing steps entailed in face perception in humans, and suggest that early stages of face processing are tuned to familiar face features.

The speed of human social interaction perception

“…The N170 response (Bentin et al, 1996) is a strongly face-selective univariate response arising around 170 ms after image onset. However, recent decoding studies have shown that many aspects of face information are represented earlier than 170 ms. For example, age, gender and identity are all decodable around 100 ms (Dobs et al, 2018). Even emotion properties like expression (100 ms (Dima et al, 2018)) and valence and arousal (150 ms (Grootswagers et al, 2017)) have been shown to come online quickly.…”

“…Second, visual recognition in primates is fast, occurring within 200ms of image onset, as expected of a largely feedforward process. These fast latencies have been demonstrated for face (Bentin et al, 1996;Dobs et al, 2018), scene (Cichy et al, 2016a;Greene and Hansen, 2018), and object (Carlson et al, 2013a;Isik et al, 2014;Yamins et al, 2014) recognition. In contrast, some visual information cannot be computed from bottom-up visual information alone.…”

Temporal Dynamics of the Neural Representation of Social Relationships

“…The results of our study show that face identity is not solely driven by low level visual properties, as captured by the HMAX model, or face shape as captured in PCA space. A recent study indicated that gender and age information is also encoded early in the MEG response to faces, especially familiar faces, and overlaps with representations of identity (Dobs et al, 2019).…”

“…In a socially interconnected world, the ability to perceive and understand complex information about other people is critical. Humans rely heavily on faces to provide social information, and we can make rapid judgments about the age, sex, trustworthiness, and identity of another within a few hundred milliseconds of viewing a face (Dobs et al, 2019;Todorov et al, 2009;Young and Burton, 2018). Recently, it has been shown that information about the social connections and network positions of those we know is represented in a distributed set of brain regions, including inferior parietal, superior temporal, and medial prefrontal cortices (Parkinson et al, 2017;Morelli et al, 2018).…”

Critical information thresholds underlying concurrent face recognition functions

“…Recently, however, a new wave of face research has emerged aimed at elucidating how the brain extracts information along different face dimensions concurrently (12)(13)(14)(15). In contrast to second-order comparisons of face functions (i.e., contrasts across different tasks/face encounters), this approach investigates the different levels of categorisation reflected in the exact same neural response elicited by a given face encounter, often by applying multivariate pattern analysis (MVPA) techniques to high temporal resolution electro/magneto-encephalographic data (EEG, MEG) (16).…”