Comparison Multi Transfer Learning Models for Deep Fake Image Recognizer
Nur Aizah Rosli, Siti Norul Huda Sheikh Abdullah, Ahmad Nazri Zamani, Anahita Ghazvini, Nor Sakinah Md Othman, Nor Alia Athirah Abdul Muariff Tajuddin · 2021
The advancement of technologies nowadays cause image digitally to become essential as an official document many available applications used for image editing. Those applications have become a threat to detect image authenticity when dealing with an abundance of digital image evidence in cyber court. Hence, many researchers realize the importance of image authentication fields as deep fake is a powerful weapon for spreading misinformation on the digital platform. Deep learning has known to obtain relevant attributes automatically in placing handcrafted features in a deep network against other single-layer networks. The objective of the research is to compare two models from Convolutional Neural Network (CNN), which are VGG19 dan ResNet50 in deep learning with transfer learning for image fake detector. A total of 1500 random images consisting of 450 forgery images from the IEEE Image Forensics Challenge in 2013 were tested on the splicing technique. From this research, we used two transfer learning techniques to identify tamper images from image splicing. Based on model VGG16 and ResNet50 transfer learning, the accuracy achieved about 94.65% and 95.08%, respectively.