Data and information dimensionality in non-cooperative face recognition

Data, information dimensionality and manifold learning techniques are related issues that are gaining prominence in biometrics. Problems dealing with large amounts of data often have dimensionality issues, leading to uncertainty and inefficiency. This chapter presents concepts of manifold learning and information geometry, and discusses how the manifold geometry can be exploited to obtain biometric data representations in lower dimensions. It is also explained how biometric data that are modeled with suitable probability distributions, can be classified accurately using geodesic distances on probabilistic manifolds, or approximations when the analytic geodesic distance solutions are not known. Also, we discuss some of the representative manifold based methods applied to face recognition, and point out future research directions.