Evrişimsel sinir ağlarına dayalı çevrimiçi yüz ırk ve cinsiyet tanıma

Facial recognition holds significant importance due to its broad implications for fairness, ethics, social justice, legal adherence, technological advancement, and public confidence. By pushing forward research efforts in this domain and crafting more precise and impartial systems, we can contribute to fostering a society that is fairer and more inclusive. Deep learning and computer vision techniques have seen widespread adoption by researchers and developers across various domains, particularly in addressing issues of bias and representation, demonstrating notable and swift progress. To distinguish individuals based on gender and race, developers often focus on analyzing colors and facial features. This thesis presents a two convolutional neural network models that identifies facial race and gender. The models underwent training, validation, testing and evaluation on datasets encompassing four racial categories (African, Asian, Indian, and Caucasian) and both genders. We curated these datasets from various sources, such as "beautiful face," "SCUT-FBP5500_v2," also known as the AFD dataset, "cnsifd_faces_bmp," "Indian_actors_faces," and "img_align_celeba." The combined datasets for the four racial groups comprise 2,400 publicly identifiable images, evenly divided into 1,200 males and 1,200 females. Each racial category includes 600 randomly selected samples, with 300 from each gender. This thesis present two models for classification of race and genders. one is multi-Class classification