Kalman Filtering for Pose-Invariant Face Recognition

We propose a novel algorithm for the identification of faces from image samples. The algorithm uses the Kalman filter to identify significant facial traits. Kalmanfaces are compact visual models that represent the invariant proportions of face classes. We employ the Kalmanfaces approach on the UMIST database, a collection of face images that were recorded under varying camera angles. Kalmanfaces show robustness against invisible facial traits and outperform the classic eigenfaces approach in terms of identification performance and algorithm speed. The paper discusses Kalmanfaces extraction, application, tunable parameters, experimental results and related work on Kalman filter application in face recognition.