Abstract
Classification images (CIs) have revealed that observers rely on a small number of pixels near the eye/brow region to discriminate faces (Sekuler et al., 2004). However, one criticism of the application of CIs in this situation is that the paradigm may allow subjects to use strategies that are not used in naturalistic situations. In particular, the methods used in the previous CI experiments may have made it easier for subjects to rely on only a small set of pixels to discriminate the stimuli. One way of addressing this problem is to use CIs to determine the pixels used to perform a same/different face discrimination task. A same/different task can potentially use a very large set of faces, and therefore make it more difficult for subjects to rely on a single feature to respond accurately. We therefore measured CIs in four subjects in a same/different face discrimination task. In our initial experiment, we used only two faces so that our results could be compared to previous CI experiments. Each trial consisted of two, 500 ms intervals separated by an ISI of 500 ms. Each interval contained a face embedded in white Gaussian noise, and the subject's task was to determine if the two faces were the same or different. Face contrast was adjusted to yield correct responses on approximately 71% of the trials. CIs were calculated from the squared, pixelwise difference between the noise fields shown on ‘same’ trials. Significant pixels were clustered near the eyes and nose. This result shows the validity of the same/different CI method for investigations of face perception, and that results from the same/different method provide similar results to those found using the standard CI paradigm. Currently, we are measuring CIs in same/different tasks that incorporate more faces, and multiple poses of the same face.