A neural-network approach for compact multi-level representation of human faces
Maxim A. Grudin · 1997
this paper is based on referencing of each new face to a freeshape face, which acts as an equivalent to experience. It contains information about reliability of localised descriptors with respect to the high-level features. Any face can be warped onto this free-shape face once positions of landmarks over the features are located. After enough experience has been learned, each new introduced model face can be warped onto the reference face in order to identify the expected positions of the most stable feature descriptors. The other descriptors are removed as being unreliable.