Gender and race classification using geodesic distance measurement

Abstract

Gender and ethnicity classifications are a long-standing challenge in the face recognition’s field. They are key-demographic traits of individuals and applied in real-world applications such as biometric and demographic research, human-computer interaction (HCI), law enforcement and online advertisements. Thus, many methods have been proposed to address gender or/and race classifications and achieved various accuracies. This research improves race and gender classification by employing a geodesic path algorithm to extract discriminative features of both gender and ethnicity. PCA is also utilized for dimensionality reduction of Gender-feature and race-feature matrices. KNN and SVM are used to classify the extracted feature. This research was tested on the face recognition technology (FERET) dataset, with classification results demonstrating high-level performance (100%) in distinguishing gender and ethnicity.