On building PCA/ICA deformable facial models

Face recognition is one of the most widely adopted fields in computational bio-metrics due to its ability to make identification while being less intrusive. Despite its success in many industrial applications, the most reliable solutions have thus far been based on an Eigenface approach. The Deformable appearance model, on the other hand, has recently attracted much interest from the Computer Vision research community due to its rather more flexible nature. As such, much effort has been made on enhancing its reliability and efficiency to on par with its renowned counterpart. Emerging ideas have been focusing on including non-linearity on not only facial boundaries but also on its texture elements, so that more realistic facial synthesis can be made. The main contribution of this paper is therefore to build on top of an AAM framework the non-linear facial appearance model with ICA. To this end, experiments on comparing with existing scheme were made, and with their pros and cons discussed. Statistical analyses suggest an optimal configuration, upon which an improved deformable face model may be built.