Eccentricity based kinship verification from facial images in the wild

Kinship verification from facial images in the wild is a promising research aiming to identify whether a facial image pair shares kinship relation by analyzing face structures. This paper proposes a novel eccentricity-based kinship verification (EKV) method to demonstrate efficacy of dominant facial sections for kinship verification. The proposed EKV method uses eccentricity of ellipse-approximated dominant facial sections as discriminative parameter to perform kinship verification. It presents two different schemes, namely single eccentricity (SE) and fused eccentricity (FE). SE scheme for EKV method employs single formulation by considering single facial section. Each selected facial section is approximated as an ellipse to compute eccentricity parameter and perform verification. Next, FE scheme for EKV method employs multiview formulation by analyzing two or more facial sections. Eccentricity of different ellipse-approximated facial sections is computed and fused to form a transformed parameter and perform verification. The proposed EKV method is demonstrated on different available kinship databases. Experimental results showcase effectiveness of EKV method with the best and competitive accuracy obtained for FE scheme on different databases.