KOMPARASI PERFORMA VGG19, RESNET50, DENSENET121, DAN MOBILENETV2 DALAM MENDETEKSI GAMBAR DEEPFAKE

ANGELINE, ANGELINE (2024) KOMPARASI PERFORMA VGG19, RESNET50, DENSENET121, DAN MOBILENETV2 DALAM MENDETEKSI GAMBAR DEEPFAKE. S1 thesis, Universitas Mercu Buana Jakarta.

[img] Text (TALK)
TALK+41519120019+ANGELINE.pdf
Restricted to Registered users only

Download (1MB)

Abstract

Deepfakes are rapidly becoming a potential cyber security threat that can manipulate images, videos and even audio so realistically that humans have difficulty distinguishing whether a piece of media is genuine or the result of manipulation by artificial intelligence. CNN is one of the methods developed as a solution. The large number of CNN variants opens up the potential for further development. The author collected from various sources 1,000 real facial images and 1,000 deepfake images which were then expanded using data augmentation technique to train, validate, and test four variants of the CNN model, namely VGG19, ResNet50, DenseNet121, and MobileNetV2, with the aim of determining the most effective variant as a base model that can be developed as deepfake detector. Performance evaluation and comparison with confusion matrix technique shows that between the four models, ResNet50 performs best with 91,5% accuracy, 90% precision, and 91,3% recall. Keywords: Convolutional Neural Networks, deepfake, image classification, VGG19, ResNet50, DenseNet121, MobileNetV2to secure other greater pleasures, or else he endures pains to avoid worse pains. Deepfake secara pesat menjadi potensi ancaman keamanan siber yang dapat memanipulasi gambar, video, bahkan audio dengan sangat realistis sehingga manusia mengalami kesulitan dalam membedakan apakah sebuah media adalah asli atau merupakan hasil manipulasi kecerdasan buatan. CNN menjadi salah satu metode yang dikembangkan sebagai solusi. Banyaknya varian model CNN membuka potensi untuk pengembangan lebih lanjut. Penulis mengumpulkan dari berbagai sumber 1,000 citra wajah asli dan 1,000 citra wajah deepfake yang kemudian diperluas dengan teknik augmentasi data untuk melatih, memvalidasi, dan menguji empat varian model CNN yaitu VGG19, ResNet50, DenseNet121, dan MobileNetV2, dengan tujuan untuk menentukan varian yang paling efektif sebagai basis model yang dapat dikembangkan menjadi detektor deepfake. Evaluasi dan perbandingan performa dengan teknik confusion matrix menunjukkan bahwa di antara keempat model, ResNet50 memiliki performa terbaik dengan akurasi 91,5%, presisi 90%, dan recall 91,3%. Kata kunci: Convolutional Neural Network, deepfake, klasifikasi gambar, VGG19, ResNet50, DenseNet121, MobileNetV2

Item Type: Thesis (S1)
Call Number CD: FIK/INFO. 24 197
Call Number: SIK/15/24/140
NIM/NIDN Creators: 41519120019
Uncontrolled Keywords: Convolutional Neural Network, deepfake, klasifikasi gambar, VGG19, ResNet50, DenseNet121, MobileNetV2
Subjects: 000 Computer Science, Information and General Works/Ilmu Komputer, Informasi, dan Karya Umum > 000. Computer Science, Information and General Works/Ilmu Komputer, Informasi, dan Karya Umum > 004 Data Processing, Computer Science/Pemrosesan Data, Ilmu Komputer, Teknik Informatika
000 Computer Science, Information and General Works/Ilmu Komputer, Informasi, dan Karya Umum > 000. Computer Science, Information and General Works/Ilmu Komputer, Informasi, dan Karya Umum > 006 Special Computer Methods/Metode Komputer Tertentu
000 Computer Science, Information and General Works/Ilmu Komputer, Informasi, dan Karya Umum > 000. Computer Science, Information and General Works/Ilmu Komputer, Informasi, dan Karya Umum > 006 Special Computer Methods/Metode Komputer Tertentu > 006.3 Artificial Intelligence/Kecerdasan Buatan > 006.32 Neural Nets (Neural Network)/Jaringan Saraf Buatan
Divisions: Fakultas Ilmu Komputer > Informatika
Depositing User: khalimah
Date Deposited: 07 Sep 2024 01:22
Last Modified: 07 Sep 2024 01:22
URI: http://repository.mercubuana.ac.id/id/eprint/91244

Actions (login required)

View Item View Item