A Pose-Invariant Face Recognition System using Linear PCMAP Model

We propose a novel pose-invariant face recognition system using a manifold representation for human faces with pose variations (linear PCMAP model) as the entry format for a database of known persons. The model's generalization capability for unknown head poses enables a continuous coverage of the pose parameter space, providing high approximation accuracy for pose estimation (analysis) and transformation (synthesis). With this model as the entry format for the database, the head pose of each known face is aligned to an arbitrary head pose of an input face, resulting in a pose-invariant recognition. Experimental results with 3D facial models recorded by a Cyberware scanner show that the recognition performance of our model against pose variations is superior to that of a single-view model and is equivalent to that of a multi-view model within a limited pose range in test samples.