Statistical image quantification of adult and infant faces

To ascertain whether there are reliable visual cues to facial differences between adult and infant faces, we performed facial shape and colour analysis for the 20 faces (10 adult and 10 infant faces) used as the training stimuli. For our facial shape analysis, all the facial landmarks placed on the 20 faces were superimposed by a generalized Procrustes analysis via tpsRelw software (version 1.65). This analysis was done to standardise face size and optimise face rotation and translation to minimise the distances between corresponding landmarks. Next, we carried out a relative warp analysis (i.e., principal component analysis (PCA)) for parameter α = 0 via tpsRelw (version 1.65) to reduce dimensionality. The first five components were selected because these components accounted for more than 95% of facial shape variation. To further identify any statistically significant differences between adult and infant facial shapes, we performed Bonferroni corrected t-tests for each of the five shape principal component (PC) scores between the adult and infant faces.

For our facial colour analysis, all the facial images were unwarped to the average face configuration across the 20 faces. Next, the RGB colour intensity of the facial images was converted into a CIELAB colour space. Then, a PCA to the LAB intensity of the facial images was carried out in MATLAB (version 2019a), and the first seven components were selected because these components accounted for more than 95% of facial colour variation. To identify any statistically significant differences between adult and infant facial coloration, we performed Bonferroni corrected t-tests for each of the seven colour PC scores between the adult and infant faces.