USER AUTHENTICATION VIA ADAPTED GENERATIVE MODELS OF FACE IMAGES

It has been previously demonstrated that systems based on local features and relatively complex generative models, namely 1D Hidden Markov Models (HMMs) and pseudo-2D HMMs, are suitable for face recognition. Recently, a simpler generative model, namely the Gaussian Mixture Model (GMM), was also shown to perform well. In most of the previous literature related to these models, the experiments were performed with controlled images (perfect face localization, controlled lighting, background, pose, expression, etc.); however, for most secure authentication applications, the system has to be robust to more challenging conditions. In this article we evaluate the performance, robustness and complexity of GMM and HMM based approaches, using both perfect and automatic face localization, on the relatively difficult BANCA database. We also evaluate different training techniques for both GMM and HMM based systems; we show that the traditionally used Maximum Likelihood (ML) training approach has problems estimating robust model parameters when there is only a few training images available; we propose to tackle this problem through the use of Maximum a Posteriori (MAP) training, where the lack of data problem can be effectively circumvented. We show that models estimated with MAP are significantly more robust and are able to generalize to adverse conditions present in the BANCA database. A positive side-effect of MAP based training is that the number of client specific parameters is less than half of the number required for ML based training. We also propose to extend the GMM approach through the use of local features with embedded positional information (hence increasing performance without sacrificing the low complexity of the approach); we show that the proposed extended GMM approach obtains performance comparable to the 1D HMM approach, while being more robust and considerably less complex. We also show that while the pseudo-2D HMM approach has overall the best performance, it requires relatively long times for training and authentication.