Get access

Modelling aspects of face processing in early infancy


  • As the scanning is a discrete procedure, it produces an image in terms of a finite matrix of pixels, each pixel representing a local area. This produced slight local sampling problems in the images. For example, even if the borders of the square-blobs were straight, in some cases the pixel corresponding to one of the corners of the square-blob was not of the same colour (black) as the square-blob. In this case, we changed the colour of this pixel, in order to get a square. Moreover, even if the square-blobs should have been identical, in some cases, one square-blob was one-pixel greater (or smaller) than the others or it was not a square because one edge was one-pixel greater than the other. We changed the colour of the concerned pixels, in order to get squares of identical dimensions. We corrected similarly the rectangular shapes forming the striped blobs of image 3. In this case, we looked at the fundamental spatial frequency of the blob, compared it to that of its position, in order to obtain the same relationship between these frequencies as in the images used by Simion. Finally, the images obtained using the scanning were not symmetrical and the shape was not identical across images 1, 2 and 3, or across images 5 and 6. We extracted a mean symmetrical shape for images 1, 2 and 3, and another for images 5 and 6 and we changed the corresponding pixels in order to obtain an identical and symmetrical shape for the images.

Address for correspondence: Francesca Acerra, INSERM U483, Université Pierre et Marie Curie, 9 quai Saint Bernard, 75252 Paris Cedex 05, France; e-mail:


The aim of this work is to understand how face processing develops from birth to 4 months by modelling some face processing abilities of infants between these ages. The neural model reproduces experimental data regarding both visual preferences at birth and right hemisphere behaviour in 4-month-old infants. The model shows that two basic properties may be sufficient to simulate face preference at birth: (a) tuning selectivity of visual neurons for spatial frequencies and (b) limited vision (CSF) of newborns. Three additional properties are sufficient to match experimental data on face processing at 4 months: (c) plasticity of lateral connections as modelled by Bayesian networks, (d) amplification of eye importance and (e) mother bias induced by the high probability of seeing mother’s face. These conditions suggest that the neural substratum for face processing could be, at birth, the retina-V1/V2-colliculus system, and at 4 months, the activation of the RH face-Fusiform area, with configural representations stored in lateral connections.