Pengaruh Augmentasi Kacamata Dan Masker Pada Pengenalan Wajah Menggunakan Deep Learning

Hidayat, Ahmad Nur (2025) Pengaruh Augmentasi Kacamata Dan Masker Pada Pengenalan Wajah Menggunakan Deep Learning. Masters thesis, Institut Teknologi Sepuluh November.

Text 6025222012-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (11MB) | Request a copy

Abstract

Proses presensi di sekolah saat ini masih dilakukan secara manual, yang memakan waktu dan rentan kesalahan. Seiring kemajuan teknologi, sistem presensi otomatis berbasis pengenalan wajah mulai dikembangkan untuk meningkatkan efisiensi dan akurasi pencatatan kehadiran siswa.
Sistem pengenalan wajah berbasis deep learning dirancang untuk mendeteksi dan mengenali identitas seseorang secara otomatis dan real-time melalui citra wajah, sehingga berpotensi menggantikan metode presensi manual yang konvensional. Namun, kehadiran aksesoris wajah seperti kacamata dan masker menjadi tantangan signifikan karena dapat mengganggu proses ekstraksi fitur wajah yang akurat, sehingga menurunkan kinerja sistem dalam kondisi nyata.
Penelitian ini mengusulkan metode augmentasi berbasis kacamata, masker, dan kombinasinya untuk memperkaya dataset. Evaluasi dilakukan melalui analisis metrik pada proses pra-pelatihan dan pasca-pelatihan, dengan membandingkan performa model pada tiga jenis dataset: data asli, data dengan augmentasi aksesoris wajah, dan data dengan augmentasi geometris. Sistem pengenalan wajah menggunakan pendekatan deep learning, dengan Sample and Computation Redistribution for Face Detection (SCRFD) untuk deteksi wajah, Additive Angular Margin Loss for Deep Face Recognition (Arcface) untuk embedding wajah, dan Multilayer Perceptron (MLP) untuk mengklasifikasikan hasil embedding ke dalam identitas yang sesuai.
Hasil penelitian menunjukkan bahwa augmentasi aksesoris wajah mampu meningkatkan akurasi sistem pengenalan wajah secara signifikan. Model yang dilatih menggunakan data wajah asli (Kategori 1) memperoleh akurasi sebesar 89%, sedangkan model dengan augmentasi aksesoris wajah (Kategori 2) mencapai akurasi tertinggi sebesar 99%, dan model dengan augmentasi geometris (Kategori 3) mencapai 91%. Peningkatan akurasi sebesar 10% dari model dasar menegaskan bahwa augmentasi aksesoris wajah lebih efektif dalam meningkatkan ketahanan model terhadap variasi penampilan nyata, seperti penggunaan masker dan kacamata, dibandingkan dengan augmentasi geometris. Hasil ini menunjukkan bahwa augmentasi yang menyerupai kondisi nyata meningkatkan akurasi sistem presensi otomatis berbasis pengenalan wajah di sekolah.
=========================================================================================================================================
The attendance process in schools is currently still done manually, which is time-consuming and error-prone. As technology advances, face recognition-based automatic attendance systems are being developed to improve the efficiency and accuracy of recording student attendance.
Deep learning-based facial recognition systems are designed to detect and recognize a person's identity automatically and in real-time through facial images, thus potentially replacing conventional manual attendance methods. However, the presence of facial accessories such as glasses and masks pose a significant challenge as they can interfere with the accurate facial feature extraction process, thus degrading the system's performance in real conditions.
This research proposes an augmentation method based on glasses, masks, and their combination to enrich the dataset. The evaluation is done through metric analysis in the pre-training and post-training process, by comparing the model performance on three types of datasets: original data, data with facial accessories augmentation, and data with geometric augmentation. The face recognition system uses a deep learning approach, with Sample and Computation Redistribution for Face Detection (SCRFD) for face detection, Additive Angular Margin Loss for Deep Face Recognition (ArcFace) for face embedding, and Multilayer Perceptron (MLP) to classify the embedded results into the corresponding identities.
The results show that facial accessories augmentation can significantly improve the accuracy of the face recognition system. The model trained using real face data (Category 1) achieved an accuracy of 89%, while the model with facial accessories augmentation (Category 2) achieved the highest accuracy of 99%, and the model with geometric augmentation (Category 3) achieved 91%. The 10% increase in accuracy from the base model confirms that facial accessories augmentation is more effective in improving the model's robustness to real appearance variations, such as the use of masks and glasses, compared to geometric augmentation. These results show that augmentation that resembles real conditions improves the accuracy of face recognition-based automatic attendance systems in schools.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	attendance system, augmentasi, deep learning, face recognition attendance, face recognition technology, video recognition. attendance system, augmentasi, deep learning, face recognition attendance, face recognition technology, video recognition.
Subjects:	A General Works > AI Indexes (General) A General Works > AI Indexes (General) T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55101-(S2) Master Thesis
Depositing User:	Ahmad Nur Hidayat
Date Deposited:	31 Jul 2025 05:51
Last Modified:	31 Jul 2025 05:51
URI:	http://repository.its.ac.id/id/eprint/123755

Actions (login required)

View Item