ANALISIS GAMBAR WAJAH PALSU: MENDETEKSI KEASLIAN GAMBAR YANG DIMANIPULASI MENGGUNAKAN METODE VARIATIONAL AUTOENCODER DAN FORENSICS DEEP NEURAL NETWORK

Abstract

Artificial intelligence (AI) is one of the technologies commonly used for automated computer systems. Artificial intelligence is designed to solve cognitive problems. The convenience provided by AI is sometimes misused, resulting in negative impacts on many people. One negative impact of AI technology misuse is deepfake. Deepfake is a technology used for image or video manipulation. The manipulation techniques used in deepfake are employed to alter images, such as faces, places, objects or even voices. Variational autoencoder (VAE) is a deep learning algorithm that can be used for facial manipulation. The result of the VAE process is an image obtained from the merging of original facial images during the training process. The new facial images generated from VAE training are called decoder images or manipulations. Both original facial images and manipulation facial images will be analyzed using the forensics deep neural network method. The analysis technique involves the use of error level analysis (ELA), which helps identify significant changes that occur in the images. Based on the testing results using both original and manipulated facial images, the applied method demonstrates the ability to detect real and fake images.