In this paper, we investigate gender classification based on 2.5D facial surface normals (facial needle-maps) which can be recovered from 2D intensity images using a non-lambertian Shape-from-shading (SFS) method. We also describe a weighted principal geodesic analysis (WPGA) method to extract features from facial surface normals. By incorporating the weight matrix into principal geodesic analysis (PGA), we control the obtained principal variance axes to be in the direction of the variance on gender information. For classification, an a posteriori probability based method is adopted. Experimental results confirms that using WPGA increases the gender discriminating power in the leading eigenvectors, and also demonstrates the feasibility of gender classification based on facial shape information.
展开▼
