Purchase this article with an account.
Matthew Peterson, Craig Abbey, Miguel Eckstein; Information distribution for face identificaiton and its relation to human strategies. Journal of Vision 2007;7(9):121. doi: https://doi.org/10.1167/7.9.121.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Introduction: Humans consistently use the visual information in the eye region of the face more than the information in the other gross features when identifying individuals (Schyns et. al., 2002; Barton et. al., 2006). Using ideal observer analysis, we previously reported that humans use the eye and mouth regions more efficiently than the nose and chin with a set of computer generated faces (Peterson et. al., 2006). However, it remains a question as to whether this disparity in efficiency is due to a recognition strategy developed through expertise of the eye region gained during years of social interaction or whether it is a result of the eye region containing better information. Here, we compare the information content of these four feature regions for identification of real faces using a Bayesian ideal observer. Methods: 70 frontal-view photographs were standardized for position, contrast and size. We then selectively masked everything except the feature region of interest. An ideal observer model compared a randomly selected template, embedded in white Gaussian noise, to all templates of the same feature and made an identification. The different feature regions were equated for area and features were aligned across templates. Results: Ideal observer performance for the eye images (71 %) was much greater than for the nose (22 %), mouth (21 %) and chin (6 %). Conclusion: The human face recognition strategy using the eye region is commensurate with the concentration of visual information in real world faces. In our previous study, this became apparent through human observers' bias to select faces based on the information in the eyes, even when there was better information in the other features. This bias, present when a novel face was displayed and the information distribution was not known, was attenuated rapidly as the observers learned which features were most informative.
This PDF is available to Subscribers Only