A prototype system for unconstrained face verification based on statistical learning

This paper discusses the video processing modules of a prototype face verification system. The main modules (face detection, registration, and verification) are all based on a feature selection plus classification pipeline that implements recently proposed statistical learning algorithms. All these modules are running on the prototype since January 2008, performing face verification in real time. Here we report quantitative results obtained both off-line, on a previously recorded dataset, and on-line, on the verification system.