Sparse coding based kinship recognition

Image based kinship recognition is a challenging problem since finding the subtle features that are reliable across a large span of ages (e.g., father and son) and gender difference (e.g., father and daughter) is very difficult. In this paper, we tackle the problem using sparse coding method, automatically classifying pairs of face images as “related” or “unrelated” (in terms of kinship). First, the face images are divided into overlapping patches. Each patch is represented by its gray values. In order to capture the spatial information of the patch, the center coordinates of the patch is used. Second, we use the sparse coding based algorithm to capture more salient properties of visual patterns than vector quantization (VQ). Here, the Locality-constrained Linear Coding (LLC) scheme is used to solve the sparse coding problem. Finally, the linear SVM classifier is used for kinship recognition since the linear SVM pays less computation complexity than nonlinear SVM. Experiments results have shown that, in terms of recognition accuracy, the suggested method outperforms the NRML [6] on the challenging database KinFace-I (KFW-I) and KinFaceW-II (KFW-II).