Kinship Verification Using a Hybrid Distance Learning Network

The estimation of kin relationships between parents and siblings based on their face image is a common biometric task conducted daily by human observers. Kin similarity is subject to significant appearance variability, as parents and siblings differ by age and gender. In this work we propose a multi-feature hybrid symmetric and asymmetric distance learning shallow network for facial kinship verification. Dual discriminative representations are learnt for the parents and their siblings using a margin maximization learning scheme, while the kin verification is formulated as a classification problem solved by SVM. 

* The work was carried out towards the PhD degree in the Faculty of Engineering, Bar-Ilan University, under the supervision of Prof. Yosi Keller