Implementasi Metode Hibrida CNN-ELM Dalam Deteksi Citra Deepfake

Abstract

The existence of Artificial Intelligence (AI) today has played a significant role in human life. In addition to bringing positive impacts, AI also has negative effects that can be detrimental to humans, one of which is Deepfake. Deepfake is the use of deep learning to forge someone's face in an image or video. This research introduces a hybrid method combining Convolutional Neural Network (CNN) and Extreme Learning Machine (ELM) to detect deepfake images. The goal of this research is to create image detection to verify the authenticity of an image in order to avoid deepfake. With the advantage of feature extraction from the CNN model and the efficient computational speed of the ELM model, the CNN-ELM hybrid method can accurately and efficiently train and test data. This research uses various scenarios to find the best parameter configuration. The results of this hybrid method achieved an average accuracy of 85.77% using 600 hidden neurons, RMSprop optimization, and ReLu activation function. This research also developed a simple GUI to allow free input of photos to verify their authenticity. This research can be one approach to detecting deepfake images.