Abstract
Suppose that you see someone for the first time through a screen window, and later see the same person through a window with a different type of screen. For image-based models of face recognition, in which faces are represented in terms of raw pixel intensity values or the outputs of wavelet filters, the altered structure of the screen in this scenario would be expected to impair recognition performance. The present experiment was designed to test this hypothesis using a sequential matching paradigm.
Each trial began with a brief presentation of a “sample” face with a neutral expression that was partially masked by a checkerboard grid of small black squares. This was followed in sequence by a pattern mask, and a new image of a “test” face that could be masked by the same checkerboard as the standard, a checkerboard that was shifted in phase by 90 or 180 degrees, or with no mask at all. The standard and test faces could depict the same or different individuals, and they could have the same or different facial expressions. Observers were required to indicate whether the two depicted individuals were the same or different by pressing an appropriate response key as quickly as possible.
The results revealed that the different masking conditions had no effect on observers' judgments, and that the overall pattern of performance (classification error and reaction times) could not be predicted from low level differences in pixel intensity values of the outputs of Gabor filters.
Supported in part by a grant from NIH.