The current study used the NIRS-adaptation paradigm to investigate the development of infants’ processing of facial identity across non-rigid facial transformation. We compared brain activation in the bilateral temporal areas between the presentation of multiple faces (different-face condition) and an identical face (same-face condition), both transformed non-rigidly by showing different facial gestures. We found that despite the non-rigid facial change, hemoglobin changes in channels around the T5 and T6 positions were significantly lower for the same-face condition than for the different-face condition only in 7- to 8-momth-olds.
The current study suggests that infants’ processing of facial identity across non-rigid facial changes develops around 7 months of age. The pattern of results we obtained in infants’ bilateral temporal region is consistent with those obtained in the fusiform face area (FFA) of adults in previous studies [15, 16]. However, this does not necessarily imply that our results reflect activity in infants’ FFA. Given that NIRS can record the hemodynamic responses only in the superficial layers of the cortex, the adaptation we showed in the present study might reflect activation not in the FFA, but in the superior temporal sulcus (STS) [8, 9].
Previous behavioral study has reported that 3- to 4-month-old infants could learn a facial identity from a movie showing non-rigid facial motion and recognize the face in expressions that differed from the learned facial image . For infants’ sensitivity to non-rigid facial transformation, ERP studies reported the ability to discriminate between an angry facial expression and a neutral expression in infants aged around 7 months [18, 19]. Infants’ sensitivity to rigid changes of face have been also reported in behavioral  and neuroimaging studies . Previous studies reported that infants’ ability to recognize face across different views was shown even in newborn infants  and becomes more robust at around 8 months after birth . Consistent with these behavioural findings, the same activity between frontal and side-view faces was reported in 8-month-old infants, not but in 5-month-old infants . Taken together, the results of these previous studies and the present study suggest that 7- to 8-month-olds develop the ability to process not only information of facial transformation, but also facial identity across the facial transformation.
The stimuli in the current study consisted of images from only five individual faces. As the images of five faces were repeatedly shown during the experiment, this might have caused attenuation in the hemodynamic responses over the course of the trials. To examine this possibility, we compared the oxy-Hb changes in the same-face and different-face conditions between the first and last trials by an ANOVA with three factors: hemisphere (right vs. left), condition (same-face vs. different-face) and trial (first vs. last trial). As a result, we found no main effect and any interaction in either 5- to 6-month-olds (main effect: hemisphere p = .23, condition p = .11, trial p = .52; interaction: hemisphere × condition p = .64, hemisphere × trial p = .74, condition × trial p = .54, hemisphere × condition × trial p = .89) or 7- to 8-month-olds (main effect: hemisphere p = .27, condition p = .84, trial p = .72; interaction: hemisphere × condition, p = .14, hemisphere × trial p = .94, condition × trial p = .74, hemisphere × condition × trial p = .41). This result suggests that increased oxy-Hb concentration for the different-face condition occurred throughout the experiment in 7- to 8-month-olds, while 5- to 6-month-olds showed no changes throughout the experiment.
In the current study, 5- to 6-month-olds did not show increases in oxy-Hb concentration in either the different-face or same-face condition. Considering previous behavioral findings that even 5-month-olds had the ability to process changes of the internal features of face [22, 23], we predicted that 5- to 6-month-olds would show increases for both conditions. The pattern of younger infants’ hemodynamic responses in this study suggests differences between behavioral responses and neuroimaging responses. Although 5-month-olds can process changes of the internal features of face, this process might not be reflected in neural responses . The 5- to 6-month-olds’ hemodynamic responses in this study are in accordance with those of our previous adaptation study examining view-invariant face processing .
Inconsistent results between this study and previous behavioral studies might also result from a difference in learning time. In this study, infants were required to recognize faces without a prior learning phase. In contrast, previous behavioral studies [22, 23] used an infant-controlled habituation procedure in which infants were given considerable time to learn the faces before the recognition test. Taken together, these studies suggest that infants aged 5 to 6 months needed sufficient learning time to process the internal features of faces. Our results with the 5- to 6-month-olds did not necessarily show that 5- to 6-month-olds lack the ability to process changes in internal facial features.
A series of our adaptation studies (the present study and Kobayashi et al. [8, 9]) have suggested that infants’ temporal areas are involved in the processing of facial identity. The size of adaptation, as indexed by the difference in activity between different-face and same-face conditions, was similar to that of our previous study for rigid facial change (Left M = 2.01, SD = 2.34, Right M = 1.79, SD = 2.07)  and to that of the present study for non-rigid facial change (Left M = 2.52, SD = 4.16, Right M = 3.24, SD = 3.59). A two-way ANOVA on the size of adaptation with hemisphere (Left vs. Right) and condition (rigid change vs. non-rigid change) revealed no difference between the two studies (main effect of condition p = .42, hemisphere p = .59, interaction p = .31). Thus, our results for small adaptation in the different-face condition did not stem from the variation of faces, e.g., the shape or location of facial features. We also have revealed that 1) the processing of facial identity invariant to size change emerges before 5 months of age, and 2) the processing of facial identity across the change, along with rigid and non-rigid facial movement, develops around 7 months of age. By applying the neural adaptation paradigm established in fMRI study to NIRS measurement in infancy, we obtained the first evidence that the processing of facial identity in the bilateral temporal areas develops from a low level to a higher level by 8 months of age.