Regression based profile face annotation from a frontal image

Statistically motivated approaches for the registration and tracking of non-rigid objects, such as the Active Appearance Model (AAM), have become increasing popular by virtue of their fast and efficient modeling and alignment, but typically they require tedious manual annotation of training images. In this paper, a regression based approach for the automatic annotation of profile face image from a single annotated frontal image is presented. This approach initially finds the correspondence between frontal and profile images with balanced graph matching, and then learns the spatial relation between scattered correspondence and the structured one. The approach is experimentally validated by automatically annotate a set of testing images with a face in arbitrary poses.