Introduction to Caricature Generation by Machine
Susan E. Brennan described a simple technique for creating caricatures in her master's thesis at the Massachusetts Institute of Technology. The basic idea is to compare the face that is to be caricatured with an average face and exaggerate the differences between the two faces. Feature vectors are the mechanism for doing this.
A feature vector is a vector that is used to represent a pattern. Normally, it extends from the origin to a point (x1, x2,..., xd) in d-dimensional feature space where x1, x2 ,..., xd are measurements of the pattern. In this application of feature vectors, the vectors extend from the point describing the average face to the point describing the inputted face. The measurements are the coordinates of various facial feature points.
A Feature Vector in 3D Feature Space
The average face is based on sets of several dozen real faces in a database of several hundred. Coordinates for each facial feature point were obtained from each face and averaged to determine the feature point coordinate for the average face. Each face has 372 measurements (since there are 186 facial feature points in 2D space).
The Average Face
This page was last updated on Thursday, April 30th, 1998.
© 1998 Ian Inc.