Purchase this article with an account.
Becky Chen, Gary Shyi; Modulation of Expression on the Generalization Gradient of Pose in Face Learning and Recognition. Journal of Vision 2017;17(10):618. doi: 10.1167/17.10.618.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
All faces start anew and how novel faces can be transformed into perceptually familiar ones, overcoming variations in lighting, pose, and expression, among others, has been an important research question awaiting for satisfactorily answers. Our previous studies have demonstrated that multiple exposures coupled with sufficient variation in either expression or pose can lead to robust recognition and significant generalization, transforming novel faces into familiar ones (Shyi & He, 2011; Shyi & Lin, 2014; Cheng & Shyi, 2014). However, whether and how expression and pose can jointly affect face learning and recognition is unknown. In the present study, we examined this issue in two experiments. In Experiment 1, using 3-D face models built from 2-D images, we largely replicated our previous findings, and showed that faces that were learned with multiple expressions can lead to better generalization in recognition than those learned with single expressions. In Experiment 2, we examined how the generalization gradient in terms of angular disparity in pose between learning and test may be modulated by facial expressions. The results revealed that happy faces yielded a gradient similar to that of neutral faces. Specifically, as angular disparity between learning and test increased, performance in recognition and generalization decreased. In contrast, sad faces yielded a relatively stable generalization gradient regardless of variation in angular disparity. Taken together, these results suggest that positive and negative expressions can have pervasively differential effects on the generalization gradient of pose in face learning and recognition.
Meeting abstract presented at VSS 2017
This PDF is available to Subscribers Only