Viés racial em algoritmos de reconhecimento facial por histogramas de gradientes orientados

Abstract

Automated face recognition systems have received extensive research and development in the past few years, and have been applied to several commercial and security applications, as well uses by police institutions. However, the reliability of the evaluation techniques over these systems have been brought into question, with analysis highlighting the existance of algorithmic bias within certain face recognition technologies. In this work, an evaluation methodology that takes demographic representativity into consideration is elaborated, which is then applied to a chosen facial recognition algorithm based on Histograms of Oriented Gradients, followed by an analysis is performed over the obtained results.