Purchase this article with an account.
Jiayu Zhan, Nicola Van Rijsbergen, Oliver Garrod, Philippe Schyns; Visualizing the Information Content of 3D Face Memory in Individual Participants. Journal of Vision 2016;16(12):211. doi: 10.1167/16.12.211.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To access the identity of a face, observers match the visual representations of the input to their face memory. Here, for the first time we reveal the 3D information content of face memory applying reverse correlation to a generative space of face identity. Using 355 faces (coded on 4735 3D vertices and 800 * 600 texture pixels) and a General Linear Model (GLM) we extracted the linear factors of ethnicity, gender and age (and their interactions). We then performed a Principal Components Analysis on the GLM residuals to extract the components of identity variance the GLM did not capture. The GLM recodes each original face as a linear combination of components of ethnicity, gender and age (and their interactions) plus a linear weighting of the 355 Principal Components of residual identity. We generated random identities with the GLM by assigning random weights to the 355 Principal Component of residual identity (S1-A). On each trial, participants (N = 10) saw 6 random identities simultaneously presented. Participants selected the random identity most similar to two familiar identities (not included in the 355 faces used to build the generative GLM) and they rated its similarity with the familiar identities (S1-A). The experiment comprised 180 blocks of 20 trials each (90 blocks per identity randomly interleaved). For each participant and identity, we linearly regressed the 3D vertice coordinates of the chosen random faces with the participant's perceived similarity ratings (S1 for details). The resulting 'Classification 3D faces' reveal faithful 3D information content representing the memory of each familiar face (S1-B), at both a local (e.g. eye size and mouth shape) and a global scale (e.g. overall face shape). For the first time, our approach and results quantify the multivariate information content of face memory, within a framework that is straightforwardly generalizable to brain measures.
Meeting abstract presented at VSS 2016
This PDF is available to Subscribers Only