Win, Vincent Daniel (2018) Pengenalan Wajah Menggunakan Convolutional Neural Network. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.
Preview |
Text
05111440000139-Undergraduate_Theses.pdf - Accepted Version Download (2MB) | Preview |
Abstract
Beberapa tahun belakangan metode deep learning menarik perhatian karena menghasilkan kinerja yang sangat baik pada beragam aplikasi. Perkembangan yang pesat ini disebabkan oleh meledaknya jumlah data dan kemampuan proses komputer.
Tugas akhir ini mengimplementasikan metode Deep Learning Convolutional Neural Network (CNN) pada klasifikasi data wajah. CNN sendiri sudah terbukti dapat mengklasifikasikan data gambar (2 dimensi spatial) dengan nilai akurasi yang tinggi, namun memerlukan dataset yang banyak. Dalam tugas akhir ini, diimplementasikan VGG16 yang merupakan sebuah arsitektur CNN, untuk pengenalan wajah pada dataset LFW dan faces94.
Dari uji coba menggunakan dataset faces94, ditemukan bahwa penggunaan fungsi aktivasi Leaky ReLU lebih baik dibandingkan ReLU dalam hal kecepatan mencapai konvergensi ketika training. Sedangkan nilai akurasi sistem dapat ditingkatkan hingga 99,6% pada dataset LFW, dengan cara melakukan augmentasi data dengan proses filter dan menerapkan face alignment.
=======================================================================================================
In recent years, deep learning has gained attention because of its flexibility in many major applications. Its recent fast development mainly attributed to the exploding growth of data availability and recent improvement in GPU technologies which utilize vector computation quickly.
This final project implements Convolutional Neural Network (CNN), a Deep Learning method mainly used to classify images, on the dataset of face images. Using CNN yield high accuracy, but generally requires vast amount of data to be effective. In this final project, VGG16 which is a CNN architecture is implemented for face recognition using the LFW and faces94 dataset.
From the testing conducted, Leaky ReLU gives better performance which resulted in faster training convergence when compared to ReLU in face recognition using faces94 dataset. Meanwhile, accuracy can be improved up to 99.6% on LFW dataset by performing data augmentation using filtering and applying face alignment.
| Item Type: | Thesis (Undergraduate) |
|---|---|
| Additional Information: | RSif 006.3 Win p-1 3100018076565 |
| Uncontrolled Keywords: | Deep Learning, Convolutional Neural Network, Pengenalan Wajah, Klasifikasi Wajah |
| Subjects: | Q Science > Q Science (General) > Q325.78 Back propagation Q Science > Q Science (General) > Q337.5 Pattern recognition systems T Technology > T Technology (General) > T57.5 Data Processing |
| Divisions: | Faculty of Information and Communication Technology > Informatics > 55201-(S1) Undergraduate Thesis |
| Depositing User: | Vincent Daniel Win |
| Date Deposited: | 10 Dec 2020 07:12 |
| Last Modified: | 10 Dec 2020 07:12 |
| URI: | http://repository.its.ac.id/id/eprint/54968 |
Actions (login required)
 |
View Item |