Face recognition and tracking using unconstrained non-linear correlation filters

Recognizing and tracking a face in a video sequence is a challenging task, specially when dealing with people and uncontrolled environments. This due to the natural variability, such as expressions, illumination, pose, occlusions, etc.. This paper propose and evaluate two strategies based on correlation for face recognition and face tracking, respectively. The proposals can be used in cascade for face tracking, first a face recognition filter is synthesized with facial regions that allow recognition of a person even when the facial image test is presented in partial form and/or contains variations in illumination, reaching approximately 95% of e_ectiveness. Then the face tracking in a video sequence, is done using an adaptive unconstrained non-linear composite filter. This filter is adapted to the changes that the face su_ers through the video sequence. Both strategies can be combined or used separately in a biometric system that allows the identification and the tracking of a person in a video sequence.