Learning multi-view deep and shallow features through new discriminative subspace for bi-subject and tri-subject kinship verification
Résumé
This paper presents the combination of deep and shallow features (multi-view features) using the proposed metric learning (SILD+WCCN/LR) approach for kinship verification. Our approach based on an automatic and more efficient two-step learning into deep/shallow information. First, five layers for deep features and five shallow features (i.e. texture and shape), representing more precisely facial features involved in kinship relations (Father-Son, Father-Daughter, Mother-Son, and Mother-Daughter) are used to train the proposed Side-Information based Linear Discriminant Analysis integrating Within Class Covariance Normalization (SILD+WCCN) method. Then, each of the features projected through the discriminative subspace of the proposed SILD+WCCN metric learning method. Finally, a Logistic Regression (LR) method is used to fuse the six scores of the projected features. To show the effectiveness of our SILD+WCNN method, we do some experiments on LFW database. In term of evaluation, the proposed automatic Facial Kinship Verification (FKV) is compared with existing ones to show its effectiveness, using two challenging kinship databases. The experimental results showed the superiority of our FKV against existing ones and reached verification rates of 86.20% and 88.59% for bi-subject matching on the KinFaceW-II and TSKinFace databases, respectively. Verification rates for tri-subject matching of 90.94% and 91.23% on the available TSKinFace database for Father-Mother-Son and Father-Mother-Daughter, respectively.