15 Detection of Augmented Facial Landmarks-based Face Swapping

Abstract

Facial images play a large part in visual communication. Such images furnish nonverbal messages such as intent, emotion and identity of a human being and have numerous applications in the fields of computer vision and graphics. Due to various image processing tools and the utilization of Deep Learning (DL) methods, it is easy even for non-experts to swap the faces in face images, aiming to construct face swap attacks in face images or videos. We focus on face swapping and its detection. Face swapping replaces a face in the destination image with a face in the source image, while maintaining image attributes. It has numerous applications, including privacy protection, computer games and entertainment. However, it could also be used for fraudulent or illicit purposes. A new method for swapping the facial images and its detection based on augmented facial landmarks and Weighted Local Magnitude Patterns (WLMP) is proposed. The 81-augmented facial landmarks are extracted and performed face swapping. The extracted WLMP features from the facial images are provided to the Support Vector Machines (SVM) classifier to detect the presence of face swap attacks. The effectiveness of the proposed approach is estimated on a real-world face image dataset. The performance of the proposed approach is evaluated with various types of SVM classifiers. The overall results show that the proposed approach effectively swaps the two facial images and detects the swapped face images from the original with accuracy of 95%.