Pose Invariant Face Recognition for New Born: Machine Learning Approach

Pose is a natural and important covariate in case of newborn and face recognition across pose can troubleshoot the approaches dealing with uncooperative subjects like newborn, in which the full power of face recognition being a passive biometric technique requires to be implemented and utilized. To handle the large pose variation in newborn, we propose a pose-adaptive similarity method that uses pose-specific classifiers to deal with different combinatorial poses. A texture based face recognition method, Speed Up Robust Feature (SURF) transform, is used to compare the descriptor of testing (probe) face with given training (gallery) face descriptor. Probes executed on the face template data of newborn described here, offer comparative benefits towards affinity for pose variations and the proposed algorithm verdicts the rank 1 accuracy of 92.1 %, which demonstrates the strength of self learning even with single training face image of newborn.