Demographics versus Biometric Automatic Interoperability

Humans are naturally experts in recognizing faces. Such skills are enforced through a mix of cultural and cognitive processes. This allows human vision system to be especially efficient and effective in processing faces in a familiar environment. Automatic recognition system are currently not able (will ever be?) to achieve similar performance, especially when cross-demographic features are involved (gender, ethnicity, and age). Recent studies suggest a significant decrease of the number of recognition errors by limiting the search space to faces with the same demographics. This can be obtained by preliminarily annotating faces with a demographic profile, or by using demographic features as soft biometrics to be determined as a support to actual recognition. Especially in the second case, a multi-demographics dataset is needed to appropriately train a recognition system, and/or to test its performance. In this paper we use EGA dataset to test how interoperability relationships between biometric and demographics can be exploited for better recognition, though avoiding human intervention to preventively select appropriate demographic parameters.