Kinship Verification through Transfer Learning
Siyu Xia, Ming Shao and Yun Fu
Due to considerable variations, i.e., poses, expressions, illumination and aging on faces, identity verification from faces is still an unsolved problem. Recent research on kinship verification imposes an even challenging problem--can we determine the kinship merely based on face images? An essential observation that faces of parent captured while they were young are more alike their children's compared with images captured when they are old is revealed by genetics research. This enlightens us the following research. First, a kinship data set named KinFace consisting of children, young parents and old parents' image pairs is collected from Internet. Second, we develop a new transfer subspace learning theory aimed at eliminating the enormous divergence of distribution between children and old parents. The key thought is to utilize some intermediate distribution close to both the source and target distribution and naturally the young parent set is suitable for this task. Through this learning process, the large gap between distributions can be significantly reduced and children-old parent verification problem becomes more discrimative. Experimental results show our assumption on the role of young parents is valid and transfer learning is effective to further enhance the verification accuracy.