Inclusão de pessoas transgênero nos algoritmos de reconhecimento automatizado de gênero para reconhecimento facial

Abstract

The techniques and applications of facial recognition are on the rise, becoming even more present in real life. Many of these applications are being created without a critical analysis of their workings and impacts on society. Automatic Gender Recognition (AGR) opens a door for technological misgendering, which affects transgender people’s life. Reasearch done on this type of technology do not include this slice of popuplation. Additionally, the AGR delivers a worse performance in comparison with the identification of cisgender individuals, as stated. As such, this peace of research analized a dataset, which has transgender people in its composition. The analysis of this dataset, was tested on pre-existent AGR algorithms, the Gender Net and analyzed the effect of this dataset with transgender people to reduce the cases of technological misgendering. Evaluation metrics for classification models were used to understand the effects of this inclusion. The analysis showed through the values of evaluation metrics that the model is efficent for really posistives classifications and relevant classifications, and when it comes to transgender people means decrease the technological misgendering problems, because these persons are present in positive and relevant classificantions.