Recognizing newborns using binocular region

Existing approaches on newborn identification focuses on recognizing them using face, inked footprints, and palm prints. While palm and inked footprints are intrusive modalities, face modality suffers from non-cooperative nature of newborns. In this research, we investigate utilization of binocular region for recognizing newborns, as this region is considered to be relatively stable in face biometrics literature. We collect a database consisting of 402 face images pertaining to 50 babies of less than 6 months of age. A set of experiments pertaining to various descriptors, including local binary patterns, dense scale invariant feature transform, and Gabor features, along with subspace learning using principal component analysis, linear discriminant analysis, and independent component analysis. Recognition performance of various approaches are compared with respect to face and binocular modalities. Verification results are reported in terms of Receiver operating characteristics curves respectively. The results show that binocular can outperform face as a modality for newborn recognition.