A novel face representation toward pose invariant face recognition

Though elastic bunch graph matching (EBGM) has a good performance on face recognition in the distortion of facial expression, it is still not robust enough to in-depth rotation. To solve this problem, a novel face representation approach based on the space-filling tree is proposed in this paper. This kind of representation shows a better performance than Elastic bunch graph matching (EBGM) in in-depth rotation of pose especially when there are only frontal images in the training set. With the proposed face representation approach, the face recognition system is built. Experimental results on the FERET standard database show that the proposed face representation approach is more effective and robust to the in-depth rotation of pose when there are only frontal images in the training set.