Abstract
It is well established that the information required to recognize a novel face is markedly different from that utilized to recognize a face that we have expertise in. However, what has yet to be addressed is the process of expertise acquisition and its effect on the utilization of the available facial information as expertise increases over time. To address this is issue, we used Bubbles (Gosselin & Schyns, 2001), a technique developed to allow categorization performance to be assigned to specific image information.
In Experiments 1 and 2, we induced expertise by presenting observers with dynamic video sequences of 10 new identities that they were instructed to learn to a criterion of 100% correct. Expertise training was given on six consecutive sessions. Following each “expertise” session observers were presented with a Bubbles paradigm task on either upright or inverted face pictures (full frontal view) different from those used in the dynamic video sequences. The experimental stimuli were computed by randomly sampling with Gaussian bubbles either a 2D image plane (Experiment 1) or a 3D space composed of a 2D image plane and 5 non overlapping spatial frequency bandwidths of one octave each, starting at 90 cycles per face (Experiment 2). Observers had to identify the individuals they had just learned through these sparse stimuli.
In both experiments, we examined the use of the available face information, revealed by the Gaussian bubbles, with first and second order analyses. First order (local) analyses examined how each bubble drove performance, whereas second order (holistic) analyses examined bubbles conjunctions. The analyses revealed 1) an increasing tendency to use the optimal local and holistic information with increasing expertise, and 2) large upright vs. inverted effects in the local and holistic information utilized