A Review of Face Recognition against Longitudinal Child Faces

It is an established fact that the face-based biometric system performance is affected by the variation that is caused as a result of aging; however, the question has not been adequately investigated for non-adults, i.e. children from birth to adulthood. The majority of research and development in automated face recognition has been focused on adults. The objective of this paper is to establish an understanding of face recognition against non-adults. This work develops a publicly available longitudinal child face database of child celebrities from images in the wild (ITWCC). This work explores the challenges of biological changes due to maturation, i.e. the face grows longer and wider, the nose expands, the lips widen, etc, i.e. craniofacial morphology, and examines the impact on face recognition. The systems chosen are: Cognitec’s FaceVacs 8.3, Open Source Biometric Recognition (SF4), principal component analysis (PCA), linear discriminant analysis (LDA), local region principal component analysis (LRPCA), and cohort linear discriminant analysis. Face matchers recorded low performance: top performance in verification is 37% TAR at 1% FAR and best rank-1 identification reached 25% recognition rate on a gallery of 301 subjects.