A Local Region-based Approach to Gender Classi.cation From Face Images

We present a novel appearance-based method for gender classification from face images. To circumvent the problem of local variations in appearance that may be caused by pose, expression, or illumination variability, we use local region analysis of the face to extract the gender classi?cation features. Given a new face image, a normalized feature vector is formed by matching N local regions of the face against some fixed set of M face images using the FaceIt algorithm, then applying the Karhunen-Loeve transform to reduce the dimensionality of this MN-dimensional vector. For the purpose of comparison, we have also implemented a holistic feature extraction method based on the well-known Eigenfaces. Gender classification is performed in a compact feature space via two standard binary classifiers; SVM and FLD. The classifier is tested via cross-validation on a database of approximately 13,000 frontal and nearly frontal face images, and the best performance of 94.2% is achieved with the local region-based feature extraction and SVM classification methods.