Horizontal image compression significantly impairs human face identity recognition

Face identity recognition (FIR) is a key function for human social interactions, supported by specialized neural mechanisms. Previous studies have reported remarkable resilience of FIR to large vertical (i.e., top to bottom) or horizontal (i.e., ear-to-ear) compression of images, generally taken as evidence that spatial (configural) cues may play a limited role in this function. However, resistance of FIR to single-axis image compression rests on limited evidence, i.e., behavioral accuracy rates obtained with repeated iconic familiar face images presented for extended response windows. Here we objectively quantify the impact of single-axis image compression on implicit neural markers of single glance FIR. Twenty-one participants were tested in a well-validated paradigm, viewing the same unfamiliar face presented 6 times per second at one of five horizontal compression levels (0%, 22%, 45%, 67%, 90%), with various identity changes occurring periodically every five faces. Robust electroencephalographic (EEG) FIR responses observed over the occipito-temporal cortex at 1.2 Hz harmonics were fully preserved at a low compression level (22%) but sharply declined with increasing compression levels, showing 40% amplitude reduction at 2/3 image compression and near-complete suppression at 90%. These observations provide original evidence against a widespread view that single-axis, in particular horizontal, image compression has little effect on human FIR. Impairment of FIR with horizontal image compression can be attributed to distortion of relative distances between features, but also to the shape of local features and of the overall face structure.