Abstract

We have reported that normally sighted observers can easily discriminate 1.5 octave bandwidth filtered face images.¹ It has been suggested that face recognition is more sensitive to the spatial frequency content of the image. We tested this hypothesis by comparing discrimination and recognition performance for decreasing bandwidth face images. Full head monochrome images of eight male and eight female models in five poses were digitized with a constant ear-to-ear distance (140 pixels). Face images were digitally filtered using bandpass filters with center frequencies at 4.0, 11.3, and 32.0 cycles/face width, and bandwidths (at full width half-maximum) of 1.5, 1.0, and 0.6 octaves with a 0.25 octave Gaussian roll-off. The faces subtended 2.7° (ear-to-ear) on the display and a linearized lookup table was used when displaying the images. Face discrimination performance was measured with a same/different experimental paradigm. Face recognition performance was measured with an eight-alternative forced choice paradigm. The data from the two paradigms were transformed to $d_E^{\text{'}}$ by Elliot's table for comparison. The results indicate that performance is the same for discrimination and recognition of 1.5 octave bandwidth filtered faces and that decreasing bandwidths decrease performance accuracy.

© 1989 Optical Society of America