Pose-invariant Face Recognition Based on a Single View*

摘要

Multi-pose face recognition always subjects to the limitation of training set, because we can't get adequate face samples to construct the face pose space or to extract effective pose-robust face features. In this paper, a novel pose-invariant face recognition algorithm based on a single view is presented assisted by 3D morphable model. First, the 3D face is reconstructed from a frontal face image, and the virtual face images with different poses can be generated by 3D face projection accordingly. Then, with adequate face pose samples, we construct the face pose space for each individual, and extract his pose-invariant face common features for recognition, which is unique compared with other works. The experimental results tested on CMU-PIE and FERET face databases are promising, and demonstrate our method is robust to face poses to a certain extent.