Análise do impacto da variação da boca e do nariz em modelos de reconhecimento facial

Abstract

Explainability in AI can help increase the reliability and transparency of these systems, improving decision-making quality. This article aims to explore the influence of mouth and nose variation on facial recognition errors in computer vision systems and propose solutions to mitigate these issues using explainable AI (XAI) techniques. It is common for these systems to exhibit detection errors, particularly in specific groups, due to bias inherent in facial recognition algorithms. To achieve the proposed objective, detailed data analyses were conducted, aaccompanied by a thorough comparison of feature similarity among different individuals, in conjunction with XAI methods were employed to clearly and comprehensibly explain the decisions made by machine learning models. In doing so, valuable insights were gained to improve the accuracy and fairness of these systems, contributing to the development of more just and reliable facial recognition systems.