Penggabungan Algoritma Dlib Face Recognition dan MTCNN untuk Mendeteksi Wajah Secara Lebih Akurat

Mahesa, Rahadian Adjie (2023) Penggabungan Algoritma Dlib Face Recognition dan MTCNN untuk Mendeteksi Wajah Secara Lebih Akurat. Other thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 05111940000221_Undergraduate_Thesis.pdf] Text
05111940000221_Undergraduate_Thesis.pdf - Accepted Version
Restricted to Repository staff only until 1 October 2025.

Download (2MB) | Request a copy

Abstract

Salah satu tantangan terbesar dalam teknik pemrosesan gambar adalah deteksi wajah, dimana tugas ini harus mengelola variasi yang dapat terjadi akibat dari perubahan cahaya, oklusi, dan fitur wajah, membuat deteksi wajah dalam gambar menjadi kompleks. Teknologi modern memungkinkan pendeteksian wajah menggunakan berbagai metode, termasuk menggunakan Pustaka Face Recognition atau Pustaka Multi-task Cascaded Convolutional Networks (MTCNN). Kedua pustaka ini memiliki keunggulan dan kelemahan masing-masing. Face Recognition memanfaatkan karakteristik wajah yang telah diajarkan sebagai input untuk memprediksi wajah dalam gambar baru, namun hal ini dapat menyebabkan wajah tidak terdeteksi dalam foto dengan karakteristik wajah yang kurang jelas. MTCNN, dengan tiga network-nya yang mendeteksi lokasi wajah dan posisi fitur wajah, membuat MTCNN lebih peka dalam pendeteksian wajah dan bisa mendeteksi false negative. Diharapkan kombinasi dari kedua pustaka ini dapat meningkatkan akurasi deteksi wajah dengan saling menutupi kekurangan masing-masing. CLAHE, rotasi foto 90 derajat, dan Face Alignment digunakan untuk ekstraksi fitur wajah, sehingga memperoleh karakteristik unik setiap wajah. Model arsitektur VGGFace digunakan untuk mendapatkan nilai dari fitur-fitur wajah tersebut dan dibandingkan perbedaan antar wajah dengan Cosine Similarity serta threshold masing-masing. Dengan threshold 0,5, hasil terbaik diperoleh dengan nilai rata-rata akurasi match wajah sebesar 98.23% dan akurasi tidak match sebesar 5.70%.
=================================================================================================================================
One of the biggest challenges in image processing techniques is face detection, where this task must manage variations that can occur due to changes in light, occlusion, and facial features, making face detection in images complex. Modern technology enables face detection using various methods, including using the Face Recognition Library or the Multi-task Cascaded Convolutional Networks (MTCNN) Library. Both of these libraries have their own advantages and disadvantages. Face Recognition utilizes taught facial characteristics as input to predict faces in new images, but this can lead to undetected faces in photos with less obvious facial characteristics. MTCNN, with its three networks that detect the location of faces and the position of facial features, makes MTCNN more sensitive in face detection and can detect false negatives. It is expected that the combination of these two libraries can improve the accuracy of face detection by covering each other's shortcomings. CLAHE, 90 degree photo rotation, and Face Alignment are used for facial feature extraction, thus obtaining the unique characteristics of each face. The VGGFace architecture model is used to obtain the values of these facial features and compare the differences between faces with Cosine Similarity and their respective thresholds. With a threshold of 0.5, the best results were obtained with an average face match accuracy of 98.23% and a non-match accuracy of 5.70%.

Item Type: Thesis (Other)
Uncontrolled Keywords: Face Recognition, MTCNN, Penggabungan, Pendeteksian Wajah, Akurasi.
Subjects: Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Q Science > QA Mathematics > QA336 Artificial Intelligence
Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Q Science > QA Mathematics > QA9.58 Algorithms
T Technology > T Technology (General) > T57.5 Data Processing
Divisions: Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User: Rahadian Adjie Mahesa
Date Deposited: 01 Dec 2023 03:39
Last Modified: 01 Dec 2023 03:39
URI: http://repository.its.ac.id/id/eprint/102856

Actions (login required)

View Item View Item