Differences in image properties across facial expressions: effects on perceived contrast, stimulus efficacy and oculomotor responses

Studies within visual psychophysics suggest that fearful facial expressions possess a special status within the visual system. This is evidenced across a range of experimental paradigms that together suggest perceptual biases for fearful expressions are driven by their low spatial frequency content. Unaddressed by this account, however, is the extent that expressions of fear compared to other faces differ in terms of the statistical image properties that define facial stimuli (their Fourier amplitude spectra), and the extent that such natural differences in image properties of expressions are influenced under experimental conditions. The present thesis contributes findings to the broader literature, demonstrating that fear expressions in particular are not naturally higher in stimulus properties known to influence salience. Broadband fear expressions instead contain lower levels of RMS contrast compared to other expressions, where this effect becomes most pronounced as their spatial frequency content increases. These expression-related differences in low-level image properties are discussed in relation to contrast normalisation and differences in perceived image contrast. Findings from Experiment 2 emphasise that facial stimuli normalised for physical contrast do not necessarily have the same perceived contrast, and that this depends on the contrast metric used to normalise faces. A fear advantage for perceived salience is significantly influenced by RMS contrast normalisation and high spatial frequency filtering. A behavioural investigation of Hedger, Adams and Garner (2015) shows that these same effects of contrast normalisation may, to some degree, inadvertently enhance the extent that fearful expressions exploit the contrast sensitivity function. Additionally, Experiments 4 and 5 explore these effects using continuous flash suppression (CFS) and saccadic latency (SL). Findings showed that CFS in particular is vulnerable to faces’ contrast content and spatial frequency content, and supports evidence of a high frequency-dependent bias for fear expressions. This contrasts with SL data where no such fear advantage, nor one that is influenced by contrast normalisation, is found when using a SL paradigm. Together, these findings are discussed in relation to the proposed perceptual bias for fearful face expressions, and the extent to which these reflect a genuine adaptive phenomena, or an inadvertent effect incurred from stimulus normalisation.