Face database (from Yale):

Brute mean face(= average of the first row) [naive way to proceed] | |

Mean face(= average of the second row) [our method] |

Modes of pure spatial deformation:

Each column stands for one mode; each column consists of one mode applied with different amplitudes (2 A, A, 0, - A, -2 A) to the mean face, where A is the standard deviation associated to this mode. The first and last lines, with amplitudes +-2A, are exaggerations in order to better visualize the variations.

Modes of pure intensity variation:

We then removed the 8th subject because of his glasses.

Modes of mixed spatial/intensity variations (click on them to see the corresponding

We also performed a

(When nothing is written, the expression has been recognized successfully; whereas a written word on a face indicates the wrong expression assignment)

Associated publications:

*Guillaume Charpiat, Olivier Faugeras and Renaud Keriven,*,**Image Statistics based on Diffeomorphic Matching***International Conference on Computer Vision***ICCV 2005**[bibtex].**More complete version here**(with the facial expression recognition task, which is joint work with Jean-Yves Audibert).

Back to the

Back to the main page