A comprehensive evaluation of feature-based AI techniques for deepfake detection

In the contemporary era, where data and information are the key source in every domain, it becomes imperative to identify, detect and distinguish between fake and authentic content available online. Recent technological innovations in the area of artificial intelligence (AI) and computer vision (CV) have been the key players both in generating and detection of these media (both images and videos). The advancement of techniques for generating, fabricating, and manipulating multimedia materials has resulted in a heightened level of realism, thereby giving rise to numerous security concerns. The adoption of these technologies has resulted in the widespread dissemination of fraudulent images and videos, which are being utilised for various criminal purposes and often featuring misleading information and impersonations of public personalities, which have detrimental consequences on the reputations of those individuals involved. The proliferation of counterfeit images poses a significant threat to national security, as these images can be exploited for the purpose of identity forgery. Therefore, it is imperative to design and develop robust algorithms for detecting counterfeit media, capable of effectively distinguishing between authentic and manipulated content. This study aims to present the generative and detection techniques of visual deep fake media using deep learning (CNN, RNN, LSTM, etc.), machine learning (SVM, KNN, Random Forest, and Decision Tree) and statistical learning (3D Morphable Model). This study also conducted an in-depth analysis of the current state of literature concerning the development and application of deepfake technology, as well as the accessibility of open-source tools for generating manipulated media. The present study provides an extensive review of face manipulation methodologies employed in the development of deep fakes, specifically focusing on Identity swap, Image Synthesis, Face Re-enactment, and Attribute Manipulation. A presented review proposed a novel taxonomy based on spatial, temporal and frequency-based features for the detection of visual Deepfake. This study out passes the existing surveys that have been presented by various other researchers in this field in terms of domain, learning methods, features and manipulation techniques used. In this study, the challenges and research gaps along with the analysis of each of these have also been presented with the intent for prioritising the development of deep fake detection tools.