An Improved DeepFake Detection Approach with NASNetLarge CNN
Ismail Ilhan, Ekrem Bali, Mehmet Karaköse
Abstract
Deep fake images are a new technology that has emerged with the development of computer vision and deep learning technologies in recent years. The development of these deep fake technologies has led to the production of many fake or manipulated products. Thus, the problem of detecting the deep fake has emerged and many methods have been developed to solve this problem. In this study, feature extraction and classification method on the dataset with NASNetLarge CNN deep learning model is proposed and a successful result is produced. In the proposed method, training and test datasets were created by removing facial regions from the video frames in the Celeb-DFv2 dataset. The architecture of the NASNetLarge model is explained and the success of the model is tested. According to the test results, an ACC value of 96.7% was obtained and compared with other methods. As a result, the study offers an easier model training with a smaller dataset than other methods and produces a competitive and successful result.