Tensor cross-view quadratic discriminant analysis for kinship verification in the wild

This paper presents a new Tensor Cross-view Quadratic Discriminant Analysis (TXQDA) method based on the XQDA method for kinship verification in the wild. Many researchers used metric learning methods and have achieved reasonably good performance in kinship verification, none of these methods looks at the kinship verification as a cross-view matching problem. To tackle this issue, we propose a tensor cross-view method to train multilinear data using local histograms of local features descriptors. Therefore, we learn a hierarchical tensor transformation to project each pair face images into the same implicit feature space, in which the distance of each positive pair is minimized and that of each negative pair is maximized. Moreover, TXQDA was proposed to separate the multifactor structure of face images (i.e. kinship, age, gender, expression, illumination and pose) from different dimensions of the tensor. Thus, our TXQDA achieves better classification results through discovering a lowdimensional tensor subspace that enlarges the margin of different kin relation classes. Experimental evaluation on five challenging databases namely Cornell KinFace, UB KinFace, TSKinFace, KinFaceW-II and FIW databases, show that the proposed TXQDA significantly outperforms the current state of the art.