Membangun Model Deep Learning untuk Pencocokon Wajah Menggunakan Ekstraksi Wajah FaceNet512

Linggar, Nethaneel Patricio (2024) Membangun Model Deep Learning untuk Pencocokon Wajah Menggunakan Ekstraksi Wajah FaceNet512. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5025201180-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (3MB) | Request a copy

Abstract

Pengembangan teknologi pengenalan wajah, khususnya melalui pendekatan deep learning seperti FaceNet512 yang mempunyai akurasi melebihi kemampuan manusia dalam dataset Labeled Faces in the Wild (LFW), telah mengubah paradigma dalam pencocokan wajah dengan memberikan representasi yang kaya fitur. Namun, isu bias dalam dataset evaluasi menyoroti terhadap ketidakseimbangan representasi, seperti banyaknya wajah orang luar dan kurangnya wajah Indonesia. Tugas akhir ini memeriksa urgensi isu tersebut dengan fokus pada evaluasi model deep learning untuk pencocokan wajah, melihat efektifitasnya menggunakan wajah asal Indonesia. Tujuan utamanya adalah melihat kinerja model ini dalam konteks sistem presensi pengenalan wajah untuk Departemen Teknik Informatika ITS. Melalui eksperimen menggunakan dataset mahasiswa, model pre-built ini terbukti akurat dan dapat diaplikasikan secara luas dengan distance metric cosine. Jika kecepatan diutamakan, detector backend OpenCV menawarkan akurasi paling tinggi dibanding detector lainnya, mendekati akurasi manusia. Namun jika kecepatan bukan sebuah aspek untuk dipertimbangkan, MTCNN memberikan performa paling tinggi, dengan satu-satunya detector dengan akurasi lebih tinggi dari manusia. Hasil Tugas Akhir ini memberikan kontribusi pada pembuktian keakuratan sebuah model yang dapat diimplementasikan dalam sistem pengenalan wajah yang mempunyai potensi aplikasi pada berbagai sistem kampus dan lokal Indonesia.
======================================================================================================================================
The development of facial recognition technology, particularly through deep learning approaches such as FaceNet512, which surpasses human accuracy on the Labeled Faces in the Wild (LFW) dataset, has revolutionized face matching by providing rich feature representations. However, bias issues in evaluation datasets highlight the imbalance in representation, such as the abundance of foreign faces and the lack of Indonesian faces. This thesis examines the urgency of this issue by focusing on the evaluation of deep learning models for face matching, assessing their effectiveness using faces of Indonesian origin. The main objective is to evaluate the performance of this model in the context of a facial recognition attendance system for the Department of Informatics Engineering at ITS. Through experiments using a student dataset, this pre-built model proved to be accurate and broadly applicable with the cosine distance metric. If speed is prioritized, the OpenCV detector backend offers the highest accuracy compared to other detectors, approaching human accuracy. However, if speed is not a concern, MTCNN provides the highest performance, being the only detector with accuracy surpassing that of humans. The results of this thesis contribute to demonstrating the accuracy of a model that can be implemented in facial recognition systems with potential applications across various campus and local Indonesian systems.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Pengenalan Wajah, Deep Learning, FaceNet512, Sistem Presensi, Face Recognition, Attendance System
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing T Technology > TA Engineering (General). Civil engineering (General) > TA1650 Face recognition. Optical pattern recognition.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Nethaneel Patricio Linggar
Date Deposited:	01 Aug 2024 05:49
Last Modified:	17 Sep 2024 03:39
URI:	http://repository.its.ac.id/id/eprint/111803

Actions (login required)

View Item