Automated Face Identification Using Volume-Based Facial Models

Face represents complex, multi dimensional, meaningful visual stimuli. Computational models for face recognition represent the problem as a high dimensional pattern recognition problem. This paper introduces an innovative facial identification method using eigenface approach on volume-based graphics rather than 2D photo-images. We propose to convert polygon mesh surface to a volumetric representation by regular sampling in a volumetric space. Our motivation is to extend existing 2D facial analysis techniques to a 3D image space by taking advantage of use of the volumetric representation. We apply principle component analysis (PCA) for dimensionality reduction. Face feature patterns are projected onto a lower dimensional PCA sub-space that spans the known facial patterns. 3D eigenface feature space is constructed for face identification.