Over the last twenty years, researchers in the field of human face recognition have tried to answer the question, how can a face be formally described? Each new face recognition system provides a different answer. In this paper, the face profile is represented using a real Fourier descriptor. Outlines of face profiles are extracted from back-lit images captured using a CCIR standard TV camera and frame-grabber. Points on the nose tip and chin are extracted and used to standardise the orientation and scale of the profile. The face curve is sampled uniformly in the vertical direction and transformed from a 2-D to a 1-D representation. After Fourier transformation, a small number of the resulting coefficients are used as a feature vector. This descriptor is insensitive to translation, scale and orientation of the original face profile. Applications to face recognition and reconstruction are illustrated and discussed in the paper.
Read full abstract