hailer-MIT/Bias-Corrected-African-Facial-Recognition

OpenCV-based face detection system trained on dark-skin datasets to reduce bias. # Bias-Corrected African Facial Recognition > ⚠️ **Notice**: This repository serves as documentation of a project I co-developed as part of a collaborative effort. The original source code was maintained in a teammate’s GitHub account, which is no longer accessible to me. As such, the current repository includes only project background, goals, and my contributions for documentation and reference purposes. ## Project Overview This project aimed to develop a facial recognition system tailored to African facial features, addressing bias in traditional facial recognition systems. Many models are trained predominantly on non-African datasets, leading to lower accuracy and fairness when applied in African contexts. ## Problem Statement Standard facial recognition models often show significant performance bias due to underrepresentation of African faces in training data. Our goal was to create a bias-corrected facial recognition system using ethically sourced African datasets to improve recognition accuracy and fairness. ## My Role in the Project As part of the development team, I contributed to: - Researching bias in facial recognition systems - Identifying and sourcing relevant African facial datasets - Designing the model architecture and preprocessing pipeline - Conducting initial testing and evaluations on model accuracy fairness ## Tools and Technologies Used - Python (NumPy, OpenCV, scikit-learn) - Dataset: African Facial Dataset (mostly collected locally with mobile phone camera) - Evaluation Metrics: Accuracy, Precision/Recall, fairness ## Code Access Status Due to a loss of contact with the original repository maintainer, I do not currently have access to the project’s full source code. This repository exists to document my work and interest in AI fairness and bias mitigation. ## Contact If you're interested in this project or working on similar topics, feel free to reach out.