Compact-Fusion Feature Framework for Ethnicity Classification

In computer vision, ethnicity classification tasks utilize images containing human faces to extract labels. Ethnicity is one of the soft biometric feature categories useful in data analysis for commercial, public, and health sectors. begins with face detection as a preprocessing process determine human’s presence; then, representation extracted from isolated facial image predict class. This study utilized four handcrafted features (multi-local binary pattern (MLBP), histogram gradient (HOG), color histogram, speeded-up-robust-features-based (SURF-based)) basis generation compact-fusion feature. The framework involves optimal selection, compact extraction, representation. final was trained tested SVM One Versus All classifier classification. When it evaluated two large datasets, UTKFace Fair Face, proposed achieved accuracy levels 89.14%, 82.19%, 73.87%, respectively, dataset or five classes Face classes. Furthermore, small number at 4790, constructed based on conventional features, competitive results compared state-of-the-art methods using deep-learning-based approach.