Fig. 1. Distribution of normal vectors on human face. (a) Computation of normal vectors; (b) 3D vector field on human face

Fig. 2. LRFE feature extraction

Fig. 3. Iterative screening of candidate points based on LRFE algorithm

Fig. 4. Schematic of human face rotation direction

Fig. 5. Divergence energy maps. (a) Expansion and shrinkage of vector field; (b) divergence map on surface; (c) divergence energy map on human face

Fig. 6. Localization results on different persons based on LRFE algorithm

Fig. 7. Relationship between iteration number and candidate point number

Fig. 8. Relationship between iteration number and running time of algorithm

Fig. 9. Pose variations in Bosphorus library

Fig. 10. Localization accuracy of N, YR30, YR45, PR and YR90 category

Fig. 11. Localization of nose tip in real scene. (a) Shape index distribution; (b) divergence distribution; (c) candidate points; (d) localization result