Purchase this article with an account.
Manabu Shikauchi, Tomohiro Shibata, Shigeyuki Oba, Shin Ishii; Neural representation of face perception in the fusiform face area. Journal of Vision 2010;10(7):667. doi: https://doi.org/10.1167/10.7.667.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Human functional magnetic resonance imaging (fMRI) studies have shown that the fusiform face area (FFA) resides at one of the highest levels of the visual pathways and is specialized for face perception (Kanwisher, et al., 1997). Although previous fMRI studies employing fMRI adaptation paradigms suggested norm-based encoding is adopted in this area (Loffler, et al., 2005; Jiang, et al., 2006), it is unclear whether reconstructing the perceived face from the fMRI signals without using the fMRI rapid adaptation paradigm is feasible, which we investigated in this study. We employed a database of photo-realistic human face images. In the fMRI experiment, participants were required to gaze at an unfamiliar face image, which is a morphed image using two face images in the database, for the target period, and to memorize it. The morphing was norm-based, based on principal component analysis (PCA) (Blanz and Vetter, 1999). After a blanking period, the two face images used for the morphed image were presented, and the participants were requested to report which face was similar to the morphed image (discrimination period). The FFA was identified by contrasting the brain activities between the target period and the blanking period, so that our analysis focused on the identified FFA regions (fROI). We found the correlative area in FFA with face variations. Face discrimination behaviors were well explained by the signal detection theory based on a face space model. We then examined whether the target face can be reconstructed from the fROI signals by using canonical correlation analysis (CCA) which finds the maximally correlated low-dimensional space between the fROI data and the target image. A good reconstruction performance in terms of the similarity between a true and reconstructed face images in the CCA space was obtained for around 30% of the trials, supporting the norm-based encoding in FFA.
This PDF is available to Subscribers Only