Electronic Theses and Dissertation

Penggunaan masker wajah sejak pandemi COVID-19 telah menimbulkan tantangan signifikan dalam sistem pengenalan wajah, karena beberapa fitur wajah menjadi tertutup dan menyebabkan penurunan akurasi. Penelitian ini bertujuan untuk menganalisis dan membandingkan kinerja dua model deep learning, yaitu VGGFace dan FaceNet512, dalam mengenali wajah dengan dan tanpa masker. Dataset yang digunakan terdiri dari data sekunder (dari Kaggle) dan data primer (pengambilan langsung), yang kemudian diproses melalui tahap pra-pemrosesan seperti resize, normalisasi, dan augmentasi data. Model dilatih menggunakan pendekatan transfer learning, dengan pengujian terhadap data validasi dan data dunia nyata. Evaluasi dilakukan menggunakan metrik akurasi, precision, recall, dan F1-score. Hasil menunjukkan bahwa model VGGFace memperoleh akurasi sebesar 93,33% pada data validasi dan 86,67% pada data primer, sedangkan FaceNet512 menghasilkan akurasi sebesar 88,33% pada data validasi dan 80,00% pada data primer. Sistem juga diimplementasikan dalam aplikasi sederhana dengan fitur anti-spoofing ke dalam file CSV. Penelitian ini memberikan kontribusi terhadap pengembangan sistem pengenalan wajah yang lebih adaptif dalam kondisi wajah tertutup masker.

The use of face masks since the COVID-19 pandemic has posed significant challenges to facial recognition systems, as several facial features are occluded, leading to decreased accuracy. This study aims to analyze and compare the performance of two deep learning models, namely VGGFace and FaceNet512, in recognizing both masked and unmasked faces. The dataset used consists of secondary data (from Kaggle) and primary data (directly collected), which were then processed through preprocessing stages such as resize, normalization, and data augmentation. The models were trained using a transfer learning approach, with testing conducted on validation data and real-world conditions. Evaluation was carried out using accuracy, precision, recall, and F1-score metrics. The results show that the VGGFace model achieved an accuracy of 93,33% on validation data and 86,67% on primary data, while FaceNet512 achieved 88,33% on validation data and 80,00% on primary data. The system was also implemented in a simple application with an anti-spoofing feature into a CSV file. This research contributes to the development of more adaptive facial recognition systems under masked face conditions.