Abstract:This study investigates the effects of occlusions on the fairness of face recognition systems, particularly focusing on demographic biases. Using the Racial Faces in the Wild (RFW) dataset and synthetically added realistic occlusions, we evaluate their effect on the performance of face recognition models trained on the BUPT-Balanced and BUPT-GlobalFace datasets. We note increases in the dispersion of FMR, FNMR, and accuracy alongside decreases in fairness according to Equilized Odds, Demographic Parity, STD of Accuracy, and Fairness Discrepancy Rate. Additionally, we utilize a pixel attribution method to understand the importance of occlusions in model predictions, proposing a new metric, Face Occlusion Impact Ratio (FOIR), that quantifies the extent to which occlusions affect model performance across different demographic groups. Our results indicate that occlusions exacerbate existing demographic biases, with models placing higher importance on occlusions in an unequal fashion, particularly affecting African individuals more severely.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **the fairness issue among different races in facial recognition systems in the presence of occlusion**. Specifically, the research aims to evaluate the impact of occlusion (such as masks, sunglasses, etc.) on the performance of facial recognition models and explore whether these occlusions exacerbate the existing demographic biases. ### Research Background and Motivation In recent years, with the increasing popularity of machine - learning - based biometric recognition applications, issues regarding the security, credibility, and potential biases of models have also attracted wide - spread attention. Researchers initially focused on improving recognition performance, but are now increasingly concerned with interpretability and bias issues. In particular, misclassification events in criminal trials and the limitations of border - control facial recognition systems have made these issues more urgent. In addition, the General Data Protection Regulation (GDPR) in the European Union stipulates that users have the right to obtain explanations, which also prompts researchers to attach more importance to the transparency and interpretability of models. ### Research Objectives 1. **Evaluate the impact of occlusion on the facial recognition performance of different racial groups**: By using the Racial Faces in the Wild (RFW) dataset and synthetic real - life occlusions, the research team evaluated the impact of occlusion on the performance of facial recognition models. 2. **Propose a new fairness evaluation method**: In order to better understand the impact of occlusion on the predictions of facial recognition models, the research team introduced a new metric - **Facial Occlusion Impact Ratio (FOIR)** - to quantify the degree of impact of occlusion on the performance of facial recognition models. 3. **Reveal how occlusion exacerbates existing biases**: The research shows that occlusion not only reduces the overall performance of the model but also leads to an exacerbation of performance differences among different races, especially having a more severe impact on individuals of African descent. ### Main Contributions - **Demonstrated the inequality of performance degradation among different races in the presence of occlusion**. - **Proposed a new method for evaluating fairness in the presence of occlusion through statistical differences**, that is, by calculating the FOIR of different sensitive populations. - **Proved that in the case of False Non - Match (FNM), the importance of occlusion for prediction varies significantly among different races**, especially African - descent samples showing a higher occlusion importance, which is consistent with the greater performance degradation of this race. ### Conclusions The research results show that occlusion not only reduces the overall performance of facial recognition systems but also exacerbates the existing demographic biases. In particular, occlusion has a disproportionate impact on the predictions of facial recognition models, especially when verifying the identities of different races. The FOIR metric proposed by the research team provides a new perspective for evaluating fairness in occlusion scenarios and is helpful for further improving the fairness and reliability of facial recognition systems in the future. --- If you have more questions or need more detailed explanations, please feel free to let me know!

Fairness Under Cover: Evaluating the Impact of Occlusions on Demographic Bias in Facial Recognition

Fairness Properties of Face Recognition and Obfuscation Systems

Evaluating Proposed Fairness Models for Face Recognition Algorithms

Rethinking Common Assumptions to Mitigate Racial Bias in Face Recognition Datasets

(Un)fair Exposure in Deep Face Rankings at a Distance

A Comprehensive Study on Face Recognition Biases Beyond Demographics

The More Secure, The Less Equally Usable: Gender and Ethnicity (Un)fairness of Deep Face Recognition along Security Thresholds

Rethinking Bias Mitigation: Fairer Architectures Make for Fairer Face Recognition

Surveying Racial Bias in Facial Recognition: Balancing Datasets and Algorithmic Enhancements

Fairness Index Measures to Evaluate Bias in Biometric Recognition

The Impact of Balancing Real and Synthetic Data on Accuracy and Fairness in Face Recognition

Explaining Bias in Deep Face Recognition via Image Characteristics

Exploring Causes of Demographic Variations In Face Recognition Accuracy

Understanding Fairness of Gender Classification Algorithms Across Gender-Race Groups

Robustness Disparities in Face Detection

Addressing Racial Bias in Facial Emotion Recognition

Investigating Bias and Fairness in Facial Expression Recognition

A Deep Dive into Dataset Imbalance and Bias in Face Identification

Score Normalization for Demographic Fairness in Face Recognition

On the Potential of Algorithm Fusion for Demographic Bias Mitigation in Face Recognition

Benchmarking Algorithmic Bias in Face Recognition: An Experimental Approach Using Synthetic Faces and Human Evaluation