Deepfake Detection using Biological Features: A Survey

Abstract

A deep learning-based technique called deepfake has made it easier to change or modify images and videos. In investigations and court, visual evidence is commonly employed. These pieces of evidence may now be a suspect due to technological advancements, particularly Deepfake. Photographs and movies that have been edited are incredibly lifelike and difficult to tell apart from the original. Deepfakes have been utilized to blackmail individuals, plan terrorist attacks, disseminate incorrect facts, defame individuals as well as foment political turmoil. Our study describes the history of the Deepfake to acquire a thorough understanding of the technology. Additionally, the study focuses on the development and detection of Deepfakes and their challenges based on physiological measurements. Here, biological features such as eyebrow recognition, eye blinking detection, eye movement detection, ear and mouth detection, and heartbeat detection are clearly described and a scope in this domain is proposed. A comparison of each biological feature concerning classifiers or techniques along with its key findings is discussed. Generated using the generative adversarial network (GANs) model, Deepfakes are created by iterating an actual data-based generation and verification task through two opposite deep learning models. It was easier to identify deepfakes by humans during the nascent stage of this technology owing to pixel collapse phenomena that generated visible artifacts in skin tone and the general face structure. However, with the technology's advancement, DeepFakes have evolved to be highly indistinguishable from natural images hence it is important to review these detection methods.