A Custom Features-Based Face Recognition using a Bag of Classifiers

Recognizing faces in unconstrained environments remains a significant challenge in face recognition research. In this paper, we present a bag of classifiers for unconstrained face recognition, to enhance recognition accuracy by leveraging the strengths of unique custom feature extraction techniques. The proposed method combines well-established facial features, including distance ratios, skin tone, texture, and a newly designed area-intensity-density feature. To ensure robust evaluation, we test the approach on three widely-used benchmark datasets—Yale, AR, and LFW, and assess performance using key metrics such as accuracy, specificity, sensitivity, F1-score, Matthews Correlation Coefficient (MCC), and Area Under the Curve (AUC). Extensive experiments demonstrate that the Artificial Neural Network (ANN) performs best on the Yale dataset, while the Random Forest (RF) classifier achieves the highest accuracy on Labeled Faces in the Wild (LFW), with results of 86.67% and 79.19%, respectively. These findings underscore the potential of the bag of classifiers approach as an effective and efficient solution for face recognition in practical applications, highlighting the importance of selecting an optimal feature extraction-classifier pair.