On developing an extended feature set for automatic face recognition

A number of novel features are studied for the automatic human face recognition. The features extracted successfully contain not only intensity but also depth features. A novel set of features describing the profile is gained from a frontal view and described by a Walsh power spectrum. The effects of rotation and illumination to the profile feature have been examined which show that they will limit the measure in practical applications but do not challenge its efficacy. Inclusion of the profile has lead to a methodology which combines the frontal and side view to add in study of face depth and protrusion. Whilst in previous studies the face contour was either ignored or described by only point measurements, in this study the face contour is described by Fourier descriptors and indicates the shape of the whole face. However the shape of this contour might be affected by hair on the forehead. Generic geometrical measurements are implemented with improved flexibility and accuracy in feature extraction algorithms. The results of assessment of these features continues to emphasise the need for an extended feature set