Challenges of Using Inconsistent Head Poses to Classify Deepfakes
Abstract
Political leaders, military leaders, and the general public gather information from images and video to make decisions. Media can be spread instantaneously throughout the world at low cost and anonymously using social media, allowing small groups to gain powerful influence. This leaves the United States vulnerable to deception by media forgery. The problem of media forgery is not new, but the recent advances in machine learning have led to the development of DeepFakes, which are more sophisticated and difficult to identify. DeepFakes are especially dangerous because they can be used to change the identity of a person in a video or image. Detecting DeepFakes has been the focus of academic research, and several techniques have been developed and show promising results on available data sets, but the suitability of these algorithms for deployment within the Department of Defense or any other critical environment is an open question. This research provides a decisive answer to this question for a promising recent analytic that uses inconsistent head poses to detect when an image is manipulated.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 2021
- Accession Number
- AD1150681
Entities
People
- Kevin D. Lutz
Organizations
- Naval Postgraduate School