Investigating the effects of gender and age group based differences in identical twins

This work investigates feature-based techniques for component-face recognition on one of the most difficult tasks: recognition of identical twins. The challenge with solving face recognition for identical twins is to find a feature extraction and template formulation approach that sufficiently separates the identical twins in match space. This work extends this premise to investigate which components of the face (eye areas, nose, or mouth) are the most discriminative features. This work uses a patch-based feature extractor and compares it to two well-known texture techniques: LBP and HOG, for the full face and the component face. We show that a single face component, the eye area, nearly matches the performance of the full face and that a simple fusion of the components outperforms the full face face on the recognition task. Further we demonstrate that the proposed feature extractor does not exhibit gender biases as does some face recognition systems, i.e. it performs almost equally on males and females. And, finally we investigate the claims that face recognition becomes easier as the twins grow older. This work adds a final contribution by experimenting on the largest public identical twins corpora available to date: ND/WVU (2009, 2010, 2011) and the CASIA Twins Face Dataset.