Facial depth forgery detection based on image gradient

With the widespread application of deep learning, many artificially generated fake images and videos appear on the Internet. However, it is difficult for people to distinguish the real from the fake ones, making the research on detecting and recognizing fake images or videos receive significant attention. Since new forgery techniques can reduce the effectiveness of specific detection methods or even make them ineffective, research on detecting facial depth forgery needs to be continuously developed. To defend against the onslaught of new facial depth forgery methods, we proposed an image gradient-based approach to transform the facial depth forgery detection problem into the recognition and analysis of video frames. Specifically, there are two key components in this approach: (1) we capture images from videos and crop the face section, which dramatically reduces the amount of data; (2) we use the image gradient operator to process the face image that extracts image features for detection and recognition. After these, we have conducted extensive experiments on different facial depth forgery datasets. Experimental results demonstrated that using our image gradient approach could effectively detect facial depth forgery and achieve excellent detection and identification performance.

Automated summary

With the widespread application of deep learning, detecting and recognizing fake images or videos has become a significant research focus. To tackle new facial depth forgery techniques, we proposed an image gradient-based approach that transforms the detection problem into video frame recognition and analysis. Our extensive experiments on various datasets demonstrated that our approach effectively detects and identifies facial depth forgery with excellent performance.