A Multi-Algorithm Feature Extraction for Accurate and Efficient Face Recognition
Muhammad Aldi Darmawan, Ivransa Zuhdi Pane, Adhi Kusnadi, Fenina Adline Twince Tobing, Rangga Winantyo, Suwito Pomalingo · 2024
Biometric technology continues to advance rapidly, with facial recognition emerging as a top priority in security and law enforcement applications. This study aims to develop a more accurate and efficient facial recognition system by combining the Discrete Cosine Transform (DCT), Gaussian Mixture Model (GMM), and Convolutional Neural Network (CNN) algorithms. This approach is designed to enhance the feature extraction process and improve classification efficiency in facial recognition systems. DCT is employed to extract low-frequency features, which are critical elements in facial patterns, while GMM is utilized to segment key facial areas, such as the eyes, nose, and mouth. The extracted features are then trained using a CNN designed with a specialized architecture. Experiments conducted on the ORL dataset achieved a peak accuracy of 97.56%, with precision and recall values of 98% each, and an F1-score of 97%. Additionally, the training time was recorded at 223.606 seconds, while the average testing time per image was 0.261 seconds. The results demonstrate that the combination of DCT, GMM, and CNN significantly outperforms conventional methods that rely solely on CNN. With high accuracy and time efficiency, this approach holds strong potential for applications in sectors such as security, forensics, and surveillance.