Implementasi Vision Transformer (ViT) untuk Klasifikasi Citra Wajah Asli dan Hasil Manipulasi Artificial Intelligence

Abstract

The rapid advancement of Artificial Intelligence (AI) has enabled the creation of highly realistic manipulated facial images, making it increasingly difficult to distinguish them from authentic images. This study aims to implement the Vision Transformer (ViT) model for classifying real and AI-generated facial images. An experimental research approach was employed using a publicly available dataset consisting of real and fake face images. The preprocessing stage included image resizing, normalization, and data augmentation to improve model generalization. The dataset was divided into training, validation, and testing subsets before the training process. The Vision Transformer Base Patch16-224 model was fine-tuned to learn the global visual features of facial images. Model performance was evaluated using accuracy, precision, recall, F1-score, confusion matrix, and ROC curve. The experimental results demonstrate that the Vision Transformer model effectively distinguishes authentic facial images from AI-generated manipulated images with high classification performance. These findings indicate that Vision Transformer is a promising approach for supporting digital image authenticity verification and enhancing digital security systems against AI-generated image manipulation.