The Mid-Band Spatial Frequency Bias in Face Recognition: A Cross-Syndrome Developmental Study of Autism and Williams Syndrome

Background: Previous research has suggested that autism may be characterised by a local bias in visuo-spatial tasks, including face processing. One possible cause of this could be a low-level bias towards high spatial frequencies. As typically-developing adults tend to rely on middle spatial frequencies for face recognition, a persistent bias towards high spatial frequencies in autism could result in reduced accuracy when recognising faces, or may represent a different strategy for achieving the same outcome. The current paradigm was adapted to use across development and involved masking different spatial frequencies in face images. Lower accuracy when a particular spatial frequency band was masked would imply that this band was used during face recognition. Objectives: The current study compared the development of spatial frequency biases in face recognition in autism, Williams syndrome and typically-developing controls. The objective of this comparison was to determine if a mid-band bias was present in the two atypical groups and whether it emerged through the same developmental processes as in the typical population. Methods: Sixty-eight children (age range = 7 years - 15 years) were grouped as being typically-developing controls (TD; N = 36), or as having high-functioning autism (HFA; N = 18) or Williams syndrome (WS; N = 14). Children learned to recognise two faces and then determined which face had been masked during presentation in a 2AFC task. Masks covered the face images at either 8, 16 or 32 cycles per image (LSF, MSF and HSF respectively). Results: The use of each spatial frequency was plotted over developmental time for the three groups. In the TD group, age significantly predicted the use of HSFs (Adjusted Rē = 0.28, p < 0.001), with 7-year-olds relying on HSF information significantly more than 15-year-olds. The use of LSFs and MSFs were not predicted by age, and an adult-like bias towards the mid-band was evident by the age of 15. Interestingly, the HFA group followed an almost identical pattern, with HSFs being used more by 7-year-olds than 15-year-olds. The WS group, however, demonstrated a greater use of LSFs at 7 years and no change in the use of HSFs with age. Both disorder groups displayed the adult-like mid-band bias found in typical development by the end of the age range studied. Conclusions: The present study has demonstrated that children with autism can be as accurate at recognising faces as typically-developing controls, and that this outcome is achieved through similar developmental changes in spatial frequency biases. Children with Williams syndrome, however, achieve the adult-like mid-band bias through a very different developmental process. These data confirm the importance of comparing syndromes across a wide age range, as a focus on adults in the current study would have found no differences between the three groups. In conclusion, any featural bias found during face recognition in high-functioning autism is not due to a persistent bias towards high spatial frequencies. Future studies should address this issue in low-functioning autism in order to compare performance with the current data.