With the rapid development of artificial intelligence, various advanced image generation and processing methods continue to emerge. In terms of fake face images, most are generated by methods based on generative adversarial networks (GANs). The increasing spread of fake face images may cause social problems such as misleading public opinion or damage to personal reputation. In this paper,we proposes a facial synthesis detection method based on deep learning and frequency domain processing. In this method, the Facenet network is used to extract facial features from images and enhance them during the extraction process. Then, fast Fourier transform (FFT) is used to analyze the extracted facial features in the frequency domain, and ResNet50 network is used to extract frequency domain features from the obtained information. Finally, the extracted facial feature vectors and image frequency domain features are fused, and the image is classified based on the obtained fusion features. This method was experimented on a 16-class dataset and achieved good detection results, compared with the optimal facial feature detection model and frequency domain feature detection model.
Fake face detection based on deep learning and frequency domain processing
2023-10-11
2602328 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Deep Neural Networks in Fake News Detection
Springer Verlag | 2022
|A Deep Learning Anomaly Detection Framework for Satellite Telemetry with Fake Anomalies
DOAJ | 2022
|Role of Machine Learning in Fake Review Detection
IEEE | 2022
|Face monitoring system based on deep learning image processing
Europäisches Patentamt | 2020
|