Fasel, Ian
Coauthors(s): Ian R. Fasel, Evan C. Smith, Javier R. Movellan <ianfasel,evan,movellan>@ucsd.edu Department of Cognitive Science University of California, San Diego
UCSD
Cognitive Science
University of California, San Diego Cognitive Science Building #139 La Jolla, CA 92093-0515
cogsci.ucsd.edu/~ianfasel

Automatic Detection of Facial Landmarks: An Exhaustive Comparison of Methods

We present an exhaustive comparison of methods for automatic detection of facial landmarks (e.g., eyes, mouth, and nose) in face images of the Feret database. The dependent variables are sensitivity (as assessed by the A' measure of the ROC curve) and posterior expected distance to the desired landmarks. The image representation methods compared include: Gabor wavelets (phase, magnitude and phase + magnitude), Haar wavelets, and Gaussian wavelets. The recognition engines include: nearest neighbor, mixtures of experts, linear regression, SNOW, support vector machines, ICA, PCA, and multilayer perceptrons. In addition we evaluate the contribution of prior distributions via extensions of Wiskott, Fellous, Kr\"{u}ger, and von der Malsburg elastic graphs approach.