Single Rotated Face Location Based on Affine Template Matching
Liang Lü · Chinese Journal of Computers · 2000
In this paper a single rotated face location algorithm based on affine template matching is presented. An average frontal face is used as the base, from which all templates of various poses are produced by affine transforms varied in stretch and pose. The templates are divided into a series of pairs: eyes in whole and face itself. The eyes in whole templates are used first to search for possible eye regions and then the face templates are matched to get possible face region, finally some mosaic rules are used to verify face regions, the region corresponding to the maximum matching is output as the result. Experiments over two test sets with 144 images and 60 images have achieved a detection rate of 79.2% and 85% respectively. Those results demonstrate the feasibility of this approach.