This work does not aim to advance the state of the art for face demographic classification systems, but rather to show how synthetic images can help tackle demographic unbalance in training them. The problem of demographic bias in both face recognition and face analysis has often been underlined in recent literature, with controversial experi- mental results. The outcomes presented here both confirm the advantage of using synthetic face images to add samples to under-represented classes and suggest that the achieved performance increase is proportional to the starting unbalance.
Using PGAN to Create Synthetic Face Images to Reduce Bias in Biometric Systems / Bozzitelli, Andrea; CAVASINNI DI BENEDETTO, Pia; DE MARSICO, Maria. - 13645:(2023), pp. 536-550. (Intervento presentato al convegno ICPR 2022 - Workshop on Understanding and Mitigating Demographic Bias in Biometric Systems (UMDBB) tenutosi a Montreal) [10.1007/978-3-031-37731-0_39].
Using PGAN to Create Synthetic Face Images to Reduce Bias in Biometric Systems
Pia Cavasinni di Benedetto;Maria De Marsico
2023
Abstract
This work does not aim to advance the state of the art for face demographic classification systems, but rather to show how synthetic images can help tackle demographic unbalance in training them. The problem of demographic bias in both face recognition and face analysis has often been underlined in recent literature, with controversial experi- mental results. The outcomes presented here both confirm the advantage of using synthetic face images to add samples to under-represented classes and suggest that the achieved performance increase is proportional to the starting unbalance.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
