Abstract
Face identification is a critical activity of daily living that may be impaired by blur or distortions caused by vision loss or prosopometamorphopsia. In this study, we examine the face inversion effect as a benchmark for understanding how distortions impact the identification of upright and inverted faces. Bandpass-filtered noise (Fpeak@1-32 cycles/face) was used to generate pixel shifts to distort calm and neutral faces from a standardized face database. The amplitude of distortion was varied using an adaptive staircase. 8 normally sighted subjects were given unlimited time to identify which of 4 distorted faces matched the identity of an undistorted reference face, each presented in a 6.5°-9.8° ellipse in a 2*2 grid. Image cues were removed from each face by equalizing them to the luminance distribution and chrominance of the average face. There was a significant interaction between face orientation and distortion frequency (F5,35 = 2.72, p = 0.0354), where sensitivity as a function of distortion frequency monotonically increased for inverted faces but was asymetrically U-shaped with a vertex at 4-8 cycles/face for upright faces. For upright faces, thresholds were lowest at mid spatial frequencies, with significant differences (p < 0.05) observed at the highest frequencies (16 and 32 cpi), supporting an asymmetric U-shaped tuning profile. In contrast, thresholds for inverted faces increased progressively across spatial frequencies, consistent with a monotonic trend. These results suggest that upright face recognition is most impacted by distortions at mid frequencies, whereas inverted face recognition declines more linearly as spatial frequency increases. The peak distortion frequency is correlated with the distance between the eyes, consistent with a critical role for eye geometry in upright face identification. These results suggest that the face inversion effect for distortion is selective for high-frequency bands.