Abstract:
A novel classification method for facial images is presented. First, the logarithmic transform of the original image is analyzed into the multiscale wavelet domain. The authors show that the subband decompositions of these images have significantly non-Gaussian statistics that are best described by families of heavy-tailed distributions such as the alpha-stable distribution. Then, the authors design an estimator that exploits these statistics. Finally, the authors compare their technique with current state-of-the-art method, based on quantification of distances computed based on identified facial landmarks and, after, they quantify the achieved performance improvement. Different kinds of trees were used as classifiers.
Some of the images we used are acquired using Kinect cameras from Microsoft, others are from the Chinese CASIA database.