The term deepfake is a portmanteau from “deep learning” and “fake”, where Artificial intelligence (AI) is used to generate synthesized human images with generative adversarial networks (GAN).
Deepfake identification is used by machine learning techniques to detect synthetic images. Machine learning plays a significant role in assisting with complicated and convoluted problems that breach human ability. The precision and consistency with which technology can reliably and easily discern the integrity of digitalized visual media are thus critical. It underlines the discernment which includes but is not limited to collecting data, analyzing specifications, as well as a literature review and project proposal.
As artificial intelligence (AI) becomes “smarter” as technology evolves the rise of deepfakes seems to accumulate. We plan on analyzing greyscale images using machine learning techniques alongside federated learning to distinguish the integrity of digitalized visual media and preserve data privacy. To justify this, residuals such as GAN-generated image fingerprints can classify an image as real or fake, remote visual photoplethysmography used for image authentication, and federated learning framework(s) for privacy preservation. The aim of the study is to distinguish synthetic deepfakes from authentic images and by maintaining data privacy.
Student:
Fergan van Jaarsveld
4164150@myuwc.ac.za
Supervisors:
Prof. Antoine Bagula
Dr. Olasupo Ajayi
Co-supervisor:
Mr. Hloniphani Maluleke