An Overview of Recent Work in Media Forensics: Methods and Threats

Abstract

In this paper, we review recent work in media forensics for digital images, video, audio (specifically speech), and documents. For each data modality, we discuss synthesis and manipulation techniques that can be used to create and modify digital media. We then review technological advancements for detecting and quantifying such manipulations. Finally, we consider open issues and suggest directions for future research.

Experts fail to reliably detect AI-generated histological data

“…It means that readers of scientific articles might easily be fooled, and that neither editors nor reviewers can rely on their personal judgment during peer review. However, though not perfect, viable technical solutions have been proposed, both for scanning scientific publications in particular (2, 24) and for the detection of image manipulation in general (12, 25, 16, 26, 27). Such forensic solutions allow both a higher throughput and consistently outperform humans if applied on the same data sets (19, 23).…”

“…Where applicable, these methods to detect fraudulent data could be complemented with procedures to authenticate the entire digital workflow including data acquisition, processing, and publication via the implementation of technical standards to secure data provenance such as C2PA (https://c2pa.org). In addition to the introduction of technical measures to promote data integrity, accessibility of the original data associated with any publication is of crucial importance for several reasons: (i) It allows access to high-quality, rather than compressed or otherwise distorted data, which removes obstacles for automated forensic methods (20), (ii) it results in images in their original format and being associated with their original meta data, the combination of which is much harder to fake than an image alone (12, 25), (iii) it is necessary to actually be able to meaningfully enforce the use of standards as C2PA, and (iv) the requirement to submit original data seems to act as a deterrence against fraudulent misconduct per se (28).…”

Non-Iterative Cluster Routing: Analysis and Implementation Strategies

“…This understanding includes recognizing when an image or video deviates from the norm in a manner that suggests manipulation, even if the specific technique used for manipulation has not been encountered by the system before. A notable application of CapsNets in this context is Capsule-Forensic, which has been shown to be effective in identifying altered or synthetically produced images and videos [17,18]. The efficacy of CapsNets in this application stems from their unique ability to encode hierarchical relationships between objects and their components, including detailed pose information.…”

GAN-Generated Faces Detection: A Survey and New Perspectives

“…DeepFakes detection involves detecting and identifying images, videos, audio, and text that have been generated or manipulated using artificial intelligence techniques [81,114,103]. DeepFakes are often created to deceive and manipulate viewers by inserting fake information into real events or spreading misinformation or creating fake news [105,92,5]. The use of DeepFakes poses a significant threat to the authenticity and credibility of social media [6], making it essential to develop reliable and effective DeepFakes detection methods [77,21,51].…”