Face Recognition Toward Dark-Skinned People

Face detection suffers by a lack of dark-skinned facial image databases. Improved facial recognition technology must overcome AI bias and promote equity. In recent years, many face recognition models using supervised learning have been constrained by the disparity between their computational demands and the quantity and composition of images representing dark-skinned individuals in the dataset, requiring model efficiency and accuracy improvements for dark-skinned races. This research addresses AI discrimination in facial recognition by creating a dataset of facial images of people of color of various ages, proposing an improved facial detection technique that better distinguishes facial features and skin tones, and implementing an advanced desktop face recognition system for diverse skin tones and age groups. This study investigates two main topics: Face detection databases with varied ethnicities, age groups, lighting conditions, and facial features are developed first: The AF dataset includes the Balanced and Fair Faces (BFF) and African celebrity face image datasets. A proposed method improves face detection by identifying facial features and skin tone. Model training and parameter optimization with TensorFlow Keras API.