DETEKSI CITRA DEEPFAKE MENGGUNAKAN CNN DENGAN INTEGRASI DILATED CONVOLUTION

Taufiqurrahman, Rahmadany Fahreza (2026) DETEKSI CITRA DEEPFAKE MENGGUNAKAN CNN DENGAN INTEGRASI DILATED CONVOLUTION. Undergraduate thesis, Universitas Pembangunan Nasional Veteran Jawa Timur.

	Text (Cover) 20081010089-cover.pdf Download (1MB)
	Text (Bab 1) 20081010089-bab1.pdf Download (189kB)
	Text (Bab 2) 20081010089-bab2.pdf Restricted to Repository staff only until 19 January 2029. Download (377kB)
	Text (Bab 3) 20081010089-bab3.pdf Restricted to Repository staff only until 19 January 2029. Download (897kB)
	Text (Bab 4) 20081010089-bab4.pdf Restricted to Repository staff only until 19 January 2029. Download (943kB)
	Text (Bab 5) 20081010089-bab5.pdf Download (173kB)
	Text (Daftar Pustaka) 20081010089-daftarpustaka.pdf Download (222kB)
	Text (Lampiran) 20081010089-lampiran.pdf Restricted to Repository staff only until 19 January 2029. Download (245kB)

Abstract

The development of artificial intelligence technology has made it possible to create fake images based on deep learning known as deepfakes. The emergence of these false images poses a serious challenge to the authenticity of digital content. One effective approach to detecting deepfake images is the Convolutional Neural Network (CNN). However, conventional CNN has limitations in reaching a wider spatial context, making it less sensitive to subtle manipulations spread across different areas of the image. To address this, this study proposes the integration of Dilated convolution into the CNN architecture. Dilated convolution expands the receptive field without significantly increasing the number of parameters, allowing the model to capture local and global features simultaneously. This study uses a dataset of real and fake facial images obtained from Kaggle with a total of 8,000 images. The model was trained using a CNN architecture with two layers of dilated convolution, average pooling, ReLu, and Dropout layers, and tested using multiple ratios of the division of the trained data and test data (50:50 to 90:10). Evaluation was conducted using a confusion matrix and performance metrics such as accuracy, precision, recall, and F1-score. The test results showed that the CNN model with dilated convolution integration was able to achieve an accuracy between 81% and 87%, with the best performance in the 90:10 scheme of 87%, and stable results in the 80:20 scheme of 85%. This performance shows an improvement over standard CNN which only reaches 81–84%. These findings prove that the addition of dilated convolution significantly improves the model's ability to capture global spatial context and detect subtle manipulations of facial imagery. Thus, the CNN-Dilated architecture has proven

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Anggraeny, Fetty Tri	NIDN0711028201	fettyanggraeny.if@upnjatim.ac.id
Thesis advisor	Sihananto, Andreas Nugroho	NIDN0012049005	Andreas.nugroho.jarkom@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA76.87 Neural computers

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

Rahmadany Fahreza Taufiqurrahman

Date Deposited:

19 Jan 2026 08:06

Last Modified:

19 Jan 2026 08:31

URI:

https://repository.upnjatim.ac.id/id/eprint/48604