Talk:Recognising a face

From A conversation about the brain
Jump to: navigation, search


I question the coarse to fine method of recognizing a face. There is a great deal of evidence that face perception relies on a relatively fixed object based channel (e.g., Gold et al., 1999[1]; Nasanen et al., 1999[2]) similar to what occurs in letter recognition (e.g., Pelli's work).

Also relevant is the work by Dakin and Watt (2009)[3] on biological bar-codes -- while Dakin and Watt use biological bar-codes and a multi-scale representation. I've used computer simulations that are identical to the bar-code computation with one exception they only use using the 10 cycles/face channel. Face perception, I believe, uses an object representation that uses the most informative spatial frequency (10 cycles/face) and orientation (horizontal) bands. This, of course, incurs a great deal of information loss, but that is supported by evidence that shows that, relative to an ideal observer, face perception is, in terms of absolute efficiency, very inefficient (< 2% of the information is used).


Chris, these points are all well taken, thank you.

The example I chose, of a person walking towards the observer, escapes from some of the issues you discuss. There may well be a peak in the useful information for face discrimination at 10 cycles per face, as these references suggest (including your work). But when the person is a long way away, there is only lower spatial frequency information available (lower in cycles/face) and still certain discriminations are possible. The same is true for very short exposure durations[4]. At closer distances, or longer exposure durations, higher frequencies become available and so finer discriminations in face-space are possible. As a comment about the information in a face that will allow it to be discriminated this is presumably uncontentious (hence, not very interesting) and it does not seem to conflict with the results you quote on the relative information content of different spatial frequencies. Instead, I was using this as an example of a situation (two situations) in which a sensory context could move from one Voronoi cell to another as the sensory information became more discriminative.


  1. Gold, J., Bennett, P. J., & Sekuler, A. B. (1999). Identification of band-pass filtered letters and faces by human and ideal observers. Vision Research, 39(21), 3537-3560.
  2. Näsänen, R. (1999). Spatial frequency bandwidth used in the recognition of facial images. Vision Research, 39(23), 3824-3833.
  3. Dakin, S. C., & Watt, R. J. (2009). Biological “bar codes” in human faces. Journal of Vision, 9(4), 2.
  4. Watt, R. J. (1987). Scanning from coarse to fine spatial scales in the human visual system after the onset of a stimulus. JOSA A, 4(10), 2006-2021.