On the Security of Image Manipulation Forensics

Abstract

Abstract. In this paper, we present a unified understanding on the formal performance evaluation for image manipulation forensics techniques. With hypothesis testing model, security is qualified as the difficulty for defeating an existing forensics system and making it generate two types of forensic errors, i.e., missing and false alarm detection. We point out that the security on false alarm risk, which is rarely addressed in current literatures, is equally significant for evaluating the performance of manipulation forensics techniques. With a case study on resampling-based composition forensics detector, both qualitative analyses and experimental results verify the correctness and rationality of our understanding on manipulation forensics security.

Systematic Review: Anti-Forensic Computer Techniques

“…Several studies [22,23,31,32] use neural networks (CNNs and GANs) to detect or generate forensic images, demonstrating their effectiveness in learning forensic traces and generating sophisticated anti-forensic attacks. Other approaches [25,26] are based on the analysis of first-and second-order statistics and Laplacian modeling to detect contrast enhancement and anti-forensic techniques.…”

“…Other approaches [25,26] are based on the analysis of first-and second-order statistics and Laplacian modeling to detect contrast enhancement and anti-forensic techniques. Methods based on CNN [22,31,32] and GAN (generative adversarial networks) [23,29] show high accuracy and robustness in detecting forgery and anti-forensic attacks. Anti-forensic approaches [25,29] focus on minimizing distortion and maintaining image quality, while others [26,27] identify anomalies introduced by anti-forensic techniques.…”

“…The integration of GANs and CNNs represents a significant advance in the detection and generation of digital image forgeries. In addition, studies that optimize anti-forensic and forensic approaches [25,32] contribute to improving the accuracy and efficiency of existing methods.…”

“…G. Cao et al [32] propose a method for evaluating the performance of forensic systems that analyze the tampering of an image. M. Fontani et al [33] propose a theoretical framework based on the Dempster-Shafer theory of evidence to merge the information provided by forensic tools and anti-forensic tools.…”

“…Both papers focus on improving the detection of manipulated images but from different perspectives. The article [32] proposes a model by which to evaluate the security of forensic techniques, considering both omission and false alarm errors. On the other hand, paper [33] focuses on countering anti-forensic techniques by fusing data from traditional forensic tools and CAF (computer-aided forensic) tools.…”